906
#* condensation approach privacy preserving data mining
#@Charu C. Aggarwal,Philip S. Yu
#t2004
#cEDBT
#index0
#id0
#%1,2,3,4,5,6,7,8,9,10
#! recent years privacy preserving data mining important problem large personal data tracked business applications cases users unwilling provide personal information unless privacy sensitive information guaranteed paper propose new framework privacy preserving data mining multi dimensional data previous work privacy preserving data mining uses perturbation approach reconstructs data distributions order perform mining approach treats dimension independently ignores correlations di erent dimensions addition requires development new distribution based algorithm data mining problem does use multi dimensional records uses aggregate distributions data input leads fundamental design data mining algorithms paper develop new exible approach privacy preserving data mining does require new problem specific algorithms maps original data set new anonymized data set anonymized data closely matches characteristics original data including correlations di erent dimensions present empirical results illustrating ectiveness method

#* design quantification privacy preserving data mining algorithms
#@Dakshi Agrawal,Charu C. Aggarwal
#t2001
#cPODS
#index1
#id1
#%
#! increasing ability track collect large amounts data use current hardware technology lead development data mining algorithms preserve user privacy recently proposed technique addresses issue privacy preservation perturbing data reconstructing distributions aggregate level order perform mining method able retain privacy accessing information implicit original attributes

#* security privacy implications data mining
#@Chris Clifton,Don Marks
#t1996
#cACM SIGMOD Workshop on Research Issues on Data Mining and Knowledge Discovery
#index2
#id2
#%
#! data mining enables discover information wedo expect databasesfi bea security privacy issue make informationavailable giving webargained position paper discusses possibleproblems solutions outlines ideas furtherresearch areafi introductiondatabase technology provides number advantagesfidata mining using automatedtools analyze corporate data help findways

#* privacy preserving association rule mining vertically partitioned data
#@Jaideep Vaidya,Chris Clifton
#t2002
#cKDD
#index3
#id3
#%1,11,12,13,14,15,16,17,18,19,20,21,22,23
#! privacy considerations constrain data mining projects paper addresses problem association rule mining transactions distributed sources site holds attributes transaction sites wish collaborate identify globally valid association rules sites reveal individual transaction data present party algorithm efficiently discovering frequent itemsets minimum support levels site revealing individual transaction values categories subject descriptors database applications data mining systems distributed databases database administrationsecurity integrity protection

#* data swapping balancing privacy precision mining logic rules
#@Vladimir Estivill-Castro,Ljiljana Brankovic
#t1999
#cDaWaK
#index4
#id4
#%
#! recent proliferation data mining tools analysis large volumes data paid little attention individual privacy issues introduce methods aimed finding balance individuals right privacy data miners need general patterns huge volumes detailed records particular focus data mining task classification decision trees base security control mechanism noise addition techniques used statis tical databases multidimensional matrix model statistical databases multidimensional cubes line analytical processing olap essentially noise addition techniques robust main drawback noise addition techniques context statistical databases low statistical quality released statistics argue data mining major requirement security control mechanism addition protect privacy ensure precise bias free statistics preserve high level descriptions knowledge constructed artificial data mining tools

#* privacy preserving mining association rules
#@Alexandre V. Evfimievski,Ramakrishnan Srikant,Rakesh Agrawal,Johannes Gehrke
#t2004
#cInf. Syst.
#index5
#id5
#%
#! present framework mining association rules fromtransactions consisting categorical items datahas randomized preserve privacy individual trans actions feasible recover association rules andpreserve privacy using straightforward uniform random ization discovered rules unfortunately exploitedto nd privacy breaches analyze nature privacybreaches propose class randomization operatorsthat eective uniform randomization inlimiting breaches derive formulae unbiasedsupport estimator variance allow cover itemset supports randomized datasets showhow incorporate formulae mining algorithms finally present experimental results validate thealgorithm applying real datasets

#* efficient approach clustering large multimedia databases noise
#@Alexander Hinneburg,Daniel A. Keim
#t1998
#cKDD
#index6
#id6
#%
#! clustering algorithms applied clustering large multimedia databases fectiveness ciency existing algorithms somewhat limited clustering multimedia databases requires cluster ing high dimensional feature vectors multimedia databases contain large amounts noisefi pa introduce new algorithm clustering large multimedia databases called denclue density based clustering fi basic idea new approach model overall point density analytically sum uence functions data pointsfi clusters identi ed determining density attractors clusters arbitrary shape easily described simple equa tion based overall density functionfi advantages new approach rm mathematical basis good clustering properties data sets large amounts noise allows compact mathematical scription arbitrarily shaped clusters high dimensional data sets signi cantly faster existing algo rithmsfi demonstrate ectiveness ciency denclue perform series experiments num ber di erent data sets cad molecular biologyfi comparison dbscan shows superiority new approachfi keywords clustering algorithms density based clus tering clustering high dimensional data clustering multimedia databases clustering presence noise introduction fast technological progress data stored databases increases fastfi types data stored come increasingly complexfi addition numerical data complex multimedia data im age cad geographic molecular biology data stored databasesfi cient retrieval com plex data usually transformed high dimensional feature vectorsfi examples feature vectors color histograms shape descriptors fourier vectors text descriptors etcfi mentioned applications databases large consist millions data objects tens hundreds dimensionsfi automated knowledge discovery large multimedia databases increasingly important research issuefi clustering trend detection databases di cult databases contain large amounts noise small portion large databases accounts clusteringfi addition known algorithms work ef ciently high dimensional data methods

#* data distortion probability distribution
#@Chong K. Liew,Uinam J. Choi,Chung J. Liew
#t1985
#cACM Trans. Database Syst.
#index7
#id7
#%
#! paper introduces data distortion probability distribution probability distortion involves steps step identify underlying density function original series estimate parameters density function second step generate series data estimated density function final step map replace generated series original replaced distorted data set probability distortion guards privacy individual belonging original data set time probability distorted series provides asymptotically statistical properties original series distribution unlike conventional point distortion probability distortion difficult compromise repeated queries provides maximum exposure statistical analysis

#* privacy interfaces information management
#@Tessa A. Lau,Oren Etzioni,Daniel S. Weld
#t1999
#cCommun. ACM
#index8
#id8
#%
#! facilitate sharing information using modern communicationnetworks users able decide aprivacy policy information conceal reveal whomfi evolution privacyinterfaces user interfaces specifying privacy policies collabclio sharing web browsinghistoriesfi experience shown privacy policiesought treated class objects policy objectsshould intensional

#* maintaining data privacy association rule mining
#@Shariq Rizvi,Jayant R. Haritsa
#t2002
#cVLDB
#index9
#id9
#%
#! data mining services require accurate input data results meaningful privacy concerns influence users provide spurious information investigate respect mining association rules users encouraged provide correct information ensuring mining process reasonable degree certainty violate privacy present scheme based probabilistic distortion user data simultaneously provide high degree privacy user retain high level accuracy mining results performance scheme validated representative real synthetic datasets

#* protecting privacy disclosing information kanonymity enforcement generalization suppression
#@Pierangela Samarati,Latanya Sweeney
#t1998
#cTechnical Report SRI-CSL-98-04
#index10
#id10
#%
#! today globally networked society places great demand dissemination sharing person specific data situations aggregate statistical information reporting norm rely heavily transfer microscopically detailed transaction encounter information happens time historically public information electronically available data linked provide electronic shadow person organization

#* mining association rules sets items large databases
#@Rakesh Agrawal,Tomasz Imielinski,Arun N. Swami
#t1993
#cSIGMOD Conference
#index11
#id11
#%
#! given large database customer transactionsfi transaction consists items purchased customer visitfi present cient algorithm generates signi association rules items databasefi algorithm incorporates bu er management novel estimation pruning techniquesfi present results applying algorithm sales data obtained large retailing company shows ectiveness algorithmfi

#* fast algorithms mining association rules
#@Rakesh Agrawal,Ramakrishnan Srikant
#t1994
#cProc. 20th Int. Conf. Very Large Data Bases, VLDB, 1994
#index12
#id12
#%
#! consider problem discovering association rulesbetween items large database sales transactions present new algorithms solving problemthat fundamentally dierent known algo rithms empirical evaluation shows algorithmsoutperform known algorithms factors ranging fromthree small problems order mag nitude large problems bestfeatures proposed algorithms combinedinto hybrid algorithm called apriorihybrid scale upexperiments apriorihybrid scales linearly withthe number transactions apriorihybrid ex cellent scale properties respect transactionsize number items database

#* privacy preserving data mining
#@Rakesh Agrawal,Ramakrishnan Srikant
#t2000
#cSIGMOD Conference
#index13
#id13
#%
#! fruitful direction future data mining research development techniques incorporate privacy concernsfi specifically address following questionfi primary task data mining development models aggregated data develop accurate models access precise information individual data records consider concrete case building decision tree classifier training data values individual records

#* extensible meta learning approach scalable accurate inductive learning
#@Philip Kin-Wah Chan,Philip Kin-Wah Chan
#t1996
#cColumbia University academic commons
#index14
#id14
#%
#! research inductive learning concentrates problems relatively small amounts data coming age ubiquitous network computing likely orders magnitude data databases available various learning problems real world importance learning algorithms assume entire data set fits main memory feasible massive amounts data especially applications data mining approach handling large data set partition data set subsets run learning algorithm subsets combine results data inherently distributed multiple sites network merging data location expensive prohibitive thesis propose investigate evaluate meta learning approach integrating results multiple learning processes approach utilizes machine learning guide integration identified main meta learning strategies combiner arbiter strategies independent learning algorithms used generating classifiers combiner strategy attempts reveal relationships learned classifiers prediction patterns arbiter strategy tries determine correct prediction classifiers different opinions various schemes strategies developed empirical results schemes obtain accurate classifiers inaccurate classifiers trained data subsets implemented analyzed schemes parallel distributed environment demonstrate scalability

#* accuracy meta learning scalable data mining
#@Philip K. Chan,Salvatore J. Stolfo
#t1997
#cJ. Intell. Inf. Syst.
#index15
#id15
#%
#! paper general approach scaling data mining applications come meta learning meta learning refers general strategy seeks learn combine number separate learning processes intelligent fashion desire meta learning architecture exhibits key behaviors meta learning strategy produce accurate final classification means meta learning architecture produce final outcome accurate conventional learning algorithm applied available data second fast relative individual sequential learning algorithm applied massive databases examples operate reasonable time paper focussed primarily issues related accuracy efficacy meta learning general strategy number empirical results presented demonstrating meta learning technically feasible wide area network computing environments

#* distributed web mining using bayesian networks multiple data streams
#@R. Chen,Krishnamoorthy Sivakumar,Hillol Kargupta
#t2001
#cICDM
#index16
#id16
#%
#! present collective approach bayesian net works distributed heterogenous web log data streams approach learn local bayesian network site using local data site identifies observations likely evidence coupling local non local variables transmits asub set observations central site bayesian network learnt central site using data transmittedfrom local site local central bayesian networks combined obtain collective bayesian net work models entire data applied techniqueto multiple data streams data centralization difficult large response time scalability issues experimental results theoretical justification demonstrate feasibility approach presented

#* efficient mining association rules distributed databases
#@David Wai-Lok Cheung,Vincent Ng,Ada Wai-Chee Fu,Yongjian Fu
#t1996
#cIEEE Trans. Knowl. Data Eng.
#index17
#id17
#%
#! sequential algorithms proposed mining association rulesfi verylittle work mining association rules distributed database direct application sequential algorithms distributed databases effective requires large amountof communication overheadfi study efficient algorithm dma proposedfi generatesa small number candidate sets requires messages support count exchangefor

#* secure multi party computation problems applications review open problems
#@Wenliang Du,Mikhail J. Atallah
#t2001
#cNSPW
#index18
#id18
#%
#! growth internet triggered tremendous opportunities cooperative computation people jointly conducting computation tasks based private inputs suppliesfi computations occur mutually untrusted parties competitorsfi example customers send remote database queries contain private information competing financial organizations jointly invest project satisfy organizations

#* secure multi party computational geometry
#@Mikhail J. Atallah,Wenliang Du
#t2001
#cWADS
#index19
#id19
#%
#! general secure multi party computation problem whenmultiple parties say alice bob private data respectively seek compute function revealingto unintended ifiefi canbe inferred knowing fi known theory thegeneral secure multi party computation problem solvable using circuitevaluation protocolsfi approach appealing

#* secure protocol computing dot products clustered distributed environments
#@Ioannis Ioannidis,Ananth Grama,Mikhail Atallah
#t2002
#cInternational Conference on Parallel Processing(ICPP'02)
#index20
#id20
#%
#! dot products form basis various applications ranging scientific computations commercial applications data mining transaction processing typical scientific computations utilizing sparse iterative solvers use repeated matrix vector products canbe viewed dot products sparse vectors database applications dot products form counting operations widespread use clustered distributed platforms operations increasingly performed networked hosts traditional apis messaging susceptible sniffing data transferred hosts compromise entire computation example domain decomposition based sparse solver entire solution reconstructed easily boundary values communicated net applications dot products performed hosts want disclose vectors need compute dot product cases need secure anonymous dot productprotocols large computational requirements underlying applications highly desirable secure protocols add minimal overhead original algorithm finally nature dot products leak limited amounts information parties detect entry party vector simply probing vector particular location zeros given constraints traditional cryptographic protocols generally unsuitable significant computational communication overheads paper present extremely efficient sufficiently secure protocol computing dot product vectors using linear algebraic techniques using analytical experimental results demonstrate superior performance terms computational overhead numerical stability security overhead party dot product computation using mpi messaging api high end workstations connected gigabit ethernet approaches multiple unsecured dot product average relative error dot products large number random normalized vectors roughly

#* privacy preserving distributed mining association rules horizontally partitioned data
#@Murat Kantarcioglu,Chris Clifton
#t2004
#cIEEE Trans. Knowl. Data Eng.
#index21
#id21
#%
#! data mining extract important knowledge fromlarge data collections collections splitamong various parties privacy concerns prevent partiesfrom directly sharing data types informationabout data paper addresses secure mining associationrules horizontally partitioned data methods incorporatecryptographic techniques minimize information shared adding little overhead mining task

#* privacy preserving association rule mining
#@
#t1000
#cEMPTY
#index22
#id22
#%
#! current trend application space systems loosely coupled dynamically bound components enables just time integration jeopardizes security information shared broker requester provider runtime particular new advances data mining knowledge discovery allow extraction hidden knowledge enormous data impose new threats seamless integration information paper consider problem building privacy preserving algorithms category data mining techniques association rule mining introduce new metrics order demonstrate security issues taken consideration general framework association rule mining complexity new heuristics similar original algorithms

#* distribution semantics new problem class distributed knowledge discovery
#@Rdiger Wirth,Michael Borth,Daimlerchrysler Ag
#t2001
#cECML01 and PKDD01
#index23
#id23
#%
#! fi research project daimlerchrysler use vehicles mobile data sources distributed knowledge discoveryfi realized current approaches suitable purposesfi aim infer global model try approximate results single joined data sourcefi treat distribution technical issue ignore distribution itself meaning models depend context derivedfi main contribution paper identification practically relevant new problem class distributed knowledge discovery addresses semantics distributionfi problem class proper framework important applications integral knowledge discovery process affecting results process itselffi outline novel solution called knowledge discovery models uses models primary input combines content driven context driven analysesfi finally discuss challenging research questions raised new problem classfi

#* transforming data satisfy privacy constraints
#@Vijay S. Iyengar
#t2002
#cKDD
#index24
#id24
#%13,25,26,27,28,29,30,31,32,33,34,35,36,37
#! data individuals entities collected widely data contain information explicitly identifies individual social security number data contain kinds personal information date birth zip code gender potentially identifying linked available data sets data shared business legal reasons paper addresses important issue preserving anonymity individuals entities data dissemination process explore preserving anonymity use generalizations suppressions potentially identifying portions data extend earlier works area various dimensions satisfying privacy constraints considered conjunction usage data disseminated allows optimize process preserving privacy specified usage particular investigate privacy transformation context data mining applications like building classification regression models second work improves previous approaches allowing exible generalizations data lastly combined thorough exploration solution space using genetic algorithm framework extensions allow transform data useful intended purpose satisfying privacy constraints general terms privacy data transformation generalization suppression predictive modeling

#* comparing sdc methods microdata basis information loss disclosure risk
#@Josep Domingo-ferrer,Josep M. Mateo-sanz,Vicen? Torra
#t2001
#cProceedings of ETK-NTTS 2001
#index25
#id25
#%
#! present paper empirical comparison sdc methodsfor microdata encompasses continuous categorical microdata basedon identification experiments try optimize tradeoff informationloss disclosure riskfi relevant sdc methods continuous categoricalmicrodata identifiedfi generic information loss measures targeted tospecific data uses defined continuous categorical casefi

#* supervised unsupervised discretization continuous features
#@James Dougherty,Ron Kohavi,Mehran Sahami
#t1995
#cICML
#index26
#id26
#%
#! supervised machine learning algo rithms require discrete feature space inthis paper review previous work tinuous feature discretization identify den ing characteristics methods duct empirical evaluation meth ods compare binning unsuperviseddiscretization method entropy based andpurity based methods supervisedalgorithms performanceof naive bayes algorithm signicantlyimproved features discretized ing entropy based method fact overthe tested datasets discretized versionof naive bayes slightly outperformed onaverage cases performance induction algo rithm signicantly improved features werediscretized advance experiments performance signicantly degraded interesting phenomenon considering thefact capable locally discretiz ing features

#* genetic algorithms search optimization machine learning
#@David E. Goldberg,John H. Holland
#t2005
#cMachine Learning
#index27
#id27
#%
#! book brings informal tutorial fashion techniques mathematical tools research results enable students practitioners apply genetic algorithms problems fields major concepts illustrated running examples major algorithms illustrated pascal programs prior knowledge gas genetics assumed minimum programming mathematics background required

#* use contextual information feature ranking discretization
#@Se June Hong
#t1997
#cIEEE Transactions on Knowledge and Data Engineering
#index28
#id28
#%
#! deriving classification rules decision trees examples important problem features discarding weak features derivation process highly desirable numeric features need discretized rule generation present new approach problems traditional techniques make use feature merits based information theoretic statistical correlation feature class instead assign merits features finding feature obligation class discrimination context features merits used rank features select feature subset discretize numeric variables experience benchmark example sets demonstrates new approach powerful alternative traditional methods paper concludes posing new technical issues arise approach

#* masking microdata files
#@Jay J. Kim,William E. Winkler,Bureau Of The Census
#t1995
#cProceedings of the Survey Research Methods Section, American Statistical Association
#index29
#id29
#%
#! government agencies collect types data confidentiality restrictions use microdata limited sworn agents working secure systems agencies restrictions severely affect public policy decisions agency access nonconfidential summary statistics necessitates creation microdata meets confidentiality requirements sufficient utility paper describes general methodology producing public use data files preserves confidentiality allows analytical uses methodology masks quantitative data using additive noise approach necessary employs reidentification swapping methodology assure confidentiality major advantages masking scheme allows obtaining precise subpopulation estimates possible known masking schemes addition controlled distortion applied prespecified subset subpopulation estimates masked file nearly identical unmasked file paper provides theoretical underpinning masking methodology results actual application using examples

#* measures disclosure risk harm
#@Diane Lambert
#t1993
#cJournal of Official Sattistics
#index30
#id30
#%
#! disclosure difficult topic definition disclosure depends context violate anonymity sensitive information revealed disclosure said occur information revealed incorrect paper tries untangle disclosure issues differentiating linking respondent record learning sensitive information linking extent released record linked respondent determines disclosure risk information revealed respondent linked released record determines disclosure harm harm wrong record identified incorrect sensitive value inferred paper measures disclosure risk harm reflect learned respondent studied implications data release policies given

#* protecting respondents identities microdata release
#@Pierangela Samarati
#t2001
#cIEEE Trans. Knowl. Data Eng.
#index31
#id31
#%
#! today globally networked society places great demand dissemination sharing information past released information tabular statistical form situations today release specific data microdata order protect anonymity entities called respondents information refers data holders remove encrypt explicit identifiers names addresses phone numbers identifying data provides guarantee anonymity released information contains data race birth date sex zip code linked publicly available information identify respondents inferring information intended disclosure paper address problem releasing microdata safeguarding anonymity respondents data refer approach based definition anonymity table provides anonymity attempts link explicitly identifying information content map information entities illustrate anonymity provided compromising integrity truthfulness information released using generalization suppression techniques introduce concept minimal generalization captures property release process distort data needed achieve anonymity present algorithm computation generalization discuss possible preference policies choose different minimal generalizations

#* protecting privacy disclosing information anonymity enforcement generalization suppression
#@Pierangela Samarati,Latanya Sweeney
#t1998
#cTechnical Report SRI-CSL-98-04, SRI Computer Science Laboratory
#index32
#id32
#%
#! today globally networked society places great demand dissemination sharing person specificdata situations aggregate statistical information reporting norm rely heavily thetransfer microscopically detailed transaction encounter informationfi happens time whenmore historically public information electronically availablefi data linkedtogether provide electronic shadow person organization

#* identification disclosure prediction disclosure microdata
#@C.J. Skinner
#t1992
#cStatistica Neerlandica
#index33
#id33
#%
#! definitions statistical disclosure identification disclosure prediction disclosure compared identification disclosure implies prediction disclosure vice versa argued sampling takes place cases prediction disclosure occurs identification disclosure does small probability present disclosure problems different normally met release aggregate statistics finally estimation population uniqueness using poisson gamma model considered

#* genitor algorithm selective pressure rank based allocation reproductive trials best
#@Darrell Whitley
#t1989
#cProceedings of the Third International Conference on Genetic Algorithms
#index34
#id34
#%
#! paper reports work past years using rank based allocation reproductive trials new evidence arguments presented suggest allocating reproductive trials according rank superior fitness proportionate reproduction ranking used slow search speed increase search speed appropriate furthermore use ranking provides degree control selective pressure possible fitness

#* statistical disclosure control practice
#@L Willenborg,T De Waal
#t1996
#cbook
#index35
#id35
#%
#! aim book discuss various aspects associated disseminating personal business data collected censuses surveys copied administrative sources problem present data form useful statistical research provide sufficient protection individuals businesses data refer major book concerned define disclosure problem deal practical circumstances

#* elements statistical disclosure control
#@L Willenborg,T De Waal
#t2001
#cbook
#index36
#id36
#%
#! statistical disclosure control discipline deals producing statistical data safe released external researchers book concentrates methodology area deals microdata individual data tabular aggregated data book attempts develop theory called paradigm statistical confidentiality modify unsafe data way safe data emerge minimum information loss book discusses safe data information loss measured modify data near optimal way decided measure safety information loss production safe data unsafe data matter solving optimization problem problems discussed book turn hard problems solved approximately authors present new results published book description area closed contrary spots awaiting fully explored indicated book book useful official social medical statisticians involved releasing personal business data statistical use operations researchers interested optimization problems involved particularly challenges present leon willenborg worked department statistical methods statistics netherlands researcher senior researcher main field research consultancy statistical disclosure control project coordinator eu funded sdc project

#* disclosure risk assessment perturbative microdata protection
#@William E. Yancey,William E. Winkler,Robert H. Creecy
#t2002
#cInference Control in Statistical Databases
#index37
#id37
#%
#! paper describes methods data perturbation include rank swapping additive noise describes enhanced methods identification using probabilistic record linkage empirical comparisons use variants framework measuring information loss identification risk introduced domingo ferrer mateo sanz

#* document centric approach static index pruning text retrieval systems
#@Stefan Buttcher,CLA Clarke
#t2006
#cCIKM
#index38
#id38
#%39,40,41,42,43,44,45,46,47,48,49,50,51
#! present static index pruning method used ad hoc document retrieval tasks follows documentcentric approach decide posting given term remain index decision based term contribution document kullback leibler divergence text collection global language model technique used decrease size index minor decrease retrieval ectiveness allows make index small fit entirely main memory single pc large text collections containing millions documents results great efficiency gains superior earlier pruning methods average response time ms gov document collection

#* pruned query evaluation using precomputed impacts
#@VO NGOC ANH,Alistair Moffat
#t2006
#cTwenty-Ninth  Annual ACM Conference on Research and Development in Information retrieval
#index39
#id39
#%
#! exhaustive evaluation ranked queries expensive particularly small subset overall ranking required queries contain common terms concern gives rise techniques dynamic query pruning methods eliminating redundant parts usual exhaustive evaluation generating demonstrably good set answers query work propose new pruning methods make use impact sorted indexes compared exhaustive evaluation new methods reduce computation performed reduce memory required accumulators reduce data transferred disk time allow performance guarantees terms precision mean average precision strong claims backed experiments using trec terabyte collection queries

#* efficient computation multiple bernoulli language model
#@Leif Azzopardi,David E. Losada
#t2006
#cECIR
#index40
#id40
#%
#! multiple bernoulli mb language model generally considered computationally expensive practical purposes superseded efficient multinomial approach model attractive properties little actually known retrieval effectiveness mb model high cost execution paper efficient implementation model achieved resulting method comparable terms efficiency standard term matching algorithms vector space model bm multinomial language model

#* efficient phrase querying auxiliary index
#@Dirk Bahle,Hugh E. Williams,Justin Zobel
#t2002
#cSIGIR
#index41
#id41
#%52,53,54,55,56,57,58,46,59,47,60,61,62,63
#! search engines need evaluate queries extremely fast challenging task given vast quantities data indexed significant proportion queries posed search engines involve phrases paper consider phrase queries efficiently supported low disk overheads previous research shown phrase queries rapidly evaluated using nextword indexes indexes twice large conventional inverted files propose combination nextword indexes inverted files solution problem experiments combined use auxiliary nextword index conventional inverted file allow evaluation phrase queries half time required evaluate queries inverted file space overhead size inverted file time savings available slight increases disk requirements general terms indexing query evaluation keywords inverted indexes nextword indexes evaluation efficiency index size stopping phrase query

#* techniques efficient query expansion
#@Bodo Billerbeck,Justin Zobel
#t2004
#cSPIRE
#index42
#id42
#%
#! fi query expansion known method improving av erage ectiveness information retrievalfi ective query expansion methods rely costly retrieval processing feed documentsfi explore alternative methods reducing query evaluation costs propose new method based keeping brief summary document memoryfi method allows query expan sion proceed times faster previously approximating ectiveness standard expansionfi

#* static index pruning information retrieval systems
#@Aya Soffer,David Carmel,Doron Cohen,Ronald Fagin,Eitan Farchi,Michael Herscovici,Yoelle S. Maarek
#t2001
#cSIGIR
#index43
#id43
#%64,65,66,67,68,69,47,70
#! introduce static index pruning methods significantly reduce index size information retrieval systems investigate uniform term based methods remove selected entries index minor effect retrieval results uniform pruning fixed cutoff threshold index entries contribution relevance scores bounded given threshold removed index term based pruning cutoff threshold determined term vary term term experimental evidence level compression term based pruning outperforms uniform pruning various measures precision present theoretical experimental evidence term based pruning scheme possible prune index greatly retrieval results good based index

#* information theoretic approach automatic query expansion
#@Claudio Carpineto,Renato de Mori,Giovanni Romano,Brigitte Bigi
#t2001
#cACM Trans. Inf. Syst.
#index44
#id44
#%
#! techniques automatic query expansion retrieved documents shown promise improving retrieval effectiveness large collections rely empirical ground shortage cross comparisons using ideas information theory present computationally simple theoretically justified method assigning scores candidate expansion terms scores used select weight expansion terms rocchio framework query reweigthing compare ranking information theoretic query expansion versus ranking query expansion techniques showing achieves better retrieval effectiveness performance measures discuss effect retrieval effectiveness main parameters involved automatic query expansion data sparseness query difficulty number selected documents number selected terms pointing interesting relationships

#* improving web search efficiency locality based static pruning method
#@Edleno Silva de Moura,Celia Francisca dos Santos,Daniel R. Fernandes,Altigran Soares da Silva,Pavel Calado,Mario A. Nascimento
#t2005
#cWWW
#index45
#id45
#%
#! unarguably fast continuous growth volume indexed indexable documents web poses great challenge search engines true regarding search ectiveness time space ciency paper present index pruning technique targeted search engines addresses issue disconsidering ect adopt new pruning strategy capable greatly reducing size search engine indices experiments using real search engine technique reduce indices stor age costs traditional lossless compres sion methods keeping loss retrieval precision minimum compared indices size compression compression rate higher eighth original size importantly results indicate reduction storage overhead query processing time reduced nearly original time loss aver age precision new method yields signi cative improve ments compared best known static prun ing method search engine indices addition technique orthogonal underlying search algorithms adopted virtually search engine

#* self indexing inverted files fast text retrieval
#@Alistair Moffat,Justin Zobel
#t1996
#cACM Trans. Inf. Syst.
#index46
#id46
#%71,72,64,73,74,75,76,77,78,79,80,81,82,83,84,85,86
#! query processing costs large text databases dominated need retrieve scan inverted list query term query response time conjunctive boolean queries informal ranked queries dramatically reduced little cost terms storage inclusion internal index inverted list method applied retrieval collection nearly million short documents experimental results selfindexing strategy adds size inverted file boolean queries terms reduce processing time fifth previous cost similarly ranked queries terms evaluated little previous time little loss retrieval ectiveness

#* filtered document retrieval frequency sorted indexes
#@Michael Persin,Justin Zobel,Ron Sacks-Davis
#t1996
#cJASIS
#index47
#id47
#%87,88,77,67,89,46,83,90,85,86
#! ranking techniques ective finding answers document collections expensive evaluate propose evaluation technique uses early recognition documents likely highly ranked reduce costs test data queries evaluated memory standard implementation degradation retrieval ectiveness cpu time disk traffic dramatically reduced designing inverted indexes explicitly support technique principle index design inverted lists sorted decreasing document frequency document number method experimentally reduces cpu time disk traffic original requirement frequency sorting lead net reduction index size regardless index compressed

#* university glasgow trec experiments web robust terabyte tracks terrier
#@Vassilis Plachouras,Ben He,Iadh Ounis
#t2004
#cProceedings of the 13th Text REtrieval Conference (TREC2004)
#index48
#id48
#%
#! participation trec test terrier modular scalable information retrieval framework tracks mixed query task web track employ decision mechanism selecting appropriate retrieval approaches query basis robust track order cope poorlyperforming queries use pre retrieval performance predictors weighting function recommender mechanism test new training approach automatic tuning term frequency normalisation parameters terabyte track employ distributed version terrier test effectiveness techniques using anchor text pseudo query expansion selecting different weighting models query

#* okapi trec
#@S.E. Robertson,S. Walker,M. Beaulieu,Peter Willett
#t2007
#cTREC-7
#index49
#id49
#%
#! article presents text retrieval research results areas automatic ad hoc retrieval information filtering vlc large collections interactive retrieval automatic ad hoc retrieval runs submitted medium title description short title run combination long run title description narrative medium short runs average precision mentioned run higher percent submitted run participant recently noticed impossibly high score topic short run led discovery mistake indexing procedures subject field la times documents indexed use field explicitly forbidden guidelines ad hoc track official runs repeated corrected index corrected results presented average precisions reduced area adaptive filtering efforts focused twin problems starting scratch assumed history relevance judgments topic having define threshold retrieval vlc runs database submitted collections area interactive retrieval pairwise comparisons okapi relevance feedback okapi okapi zprise

#* compression inverted indexes fast query evaluation
#@Falk Scholer,Hugh E. Williams,John Yiannis,Justin Zobel
#t2002
#cSIGIR
#index50
#id50
#%46,47,60,62
#! compression reduces size indexes time needed evaluate queries paper revisit compression inverted lists document postings store position frequency indexed terms considering approaches improving retrieval efficiency better implementation better choice integer compression schemes propose simple optimisations known integer compression schemes experimentally lead significant reductions time second explore impact choice compression scheme retrieval efficiency experiments large collections data surprising results use simple byte aligned codes halves query evaluation time compared compact golomb rice bitwise compression schemes index fits entirely memory byte aligned codes result faster query evaluation does uncompressed index emphasising cost transferring data memory cpu cache appropriately compressed index uncompressed index byte aligned schemes modest space overhead compact schemes result indexes size collection byte aligned scheme conclude fast byte aligned codes used store integers inverted lists

#* query evaluation strategies optimization
#@Howard Turtle,James Flood
#t1995
#cInformation Processing & Management
#index51
#id51
#%
#! discusses query evaluation strategies used large text retrieval systems term time document time describes optimization techniques reduce query evaluation costs presents simulation results compare performance optimization techniques applied natural language query evaluation

#* vector space ranking ective early termination
#@Vo Ngoc Anh,Owen de Kretser,Alistair Moffat
#t2001
#cAnnual ACM Conference on Research and Development in Information Retrieval
#index52
#id52
#%
#! considerable research effort invested improving effectiveness information retrieval systems techniques relevance feedback thesaural expansion pivoting provide better quality responses queries tested standard evaluation frameworks enhancements add cost evaluating queries paper consider pragmatic issue improve cost effectiveness searching new inverted file structure using quantized weights provides superior retrieval effectiveness compared conventional inverted file structures early termination heuristics employed able reach similar effectiveness levels computational cost provide better cost performance compromise previous inverted file organisations

#* compaction techniques nextword indexes
#@Dirk Bahle,Hugh E. Williams,Justin Zobel
#t2001
#cSPIRE
#index53
#id53
#%
#! queries text search engines ranked boolean phrase querying powerful technique ning searches expensive implement conventional indexes work nextword index proposed structure speci cally designed phrase queries nextword indexes relatively large paper introduce new compaction techniques nextword indexes contrast index compression schemes techniques lossy allow resolution phrase queries false match checking experimentally novel techniques lead signi savings index size

#* optimised phrase querying browsing text databases
#@Dirk Bahle,Hugh E. Williams,Justin Zobel
#t2001
#cProceedings of the Australasian Computer Science Conference
#index54
#id54
#%
#! search systems querying large document collectioneor example web search enginesare based understood information retrieval principles systems eficient effective finding answers user information needs expressed informal ranked structured boolean queries phrase querying browsing additional techniques augment replace conventional querying tools paper propose optimisations phrase quelying nextword index eficient structure phrase based searching careful consideration search terms evaluated query plan optimisation order evaluation plan reduce query evaluation costs factor jive conclude phrase querying browsing nextword indexes ordered query plan used browsing querying optimised phrase querying practical large text collections

#* interactive internet search keyword directory query reformulation mechanisms compared
#@Peter Bruza,Robert McArthur,Simon Dennis
#t2000
#cSIGIR
#index55
#id55
#%
#! article compares search effectiveness usingquery based internet search google searchengine directory based search yahoo phrasebasedquery reformulation assisted search hyperindex browser means controlled userbasedexperimental studyfi focus evaluateaspects search processfi cognitive load wasmeasured using secondary digit monitoring task toquantify effort user various search states independent

#* relevance ranking term queries
#@CLA Clarke,Gordon V. Cormack,EA Tudhope
#t2000
#cInformation Processing and Management
#index56
#id56
#%
#! search systems querying large document collectioneor example web search enginesare based understood information retrieval principles systems eficient effective finding answers user information needs expressed informal ranked structured boolean queries phrase querying browsing additional techniques augment replace conventional querying tools paper propose optimisations phrase quelying nextword index eficient structure phrase based searching careful consideration search terms evaluated query plan optimisation order evaluation plan reduce query evaluation costs factor jive conclude phrase querying browsing nextword indexes ordered query plan used browsing querying optimised phrase querying practical large text collections

#* phrase recognition expansion short precision biased queries based query log
#@Erika F. de Lima,Jan O. Pedersen
#t1999
#cSIGIR
#index57
#id57
#%
#! paper examine question query parsing world wide web queries present novel method phrase recognition expansionfi given training corpus approximately million web queries handwritten context free grammar em algorithm used estimate parameters probabilistic context free grammar pcfg developed carroll fi use pcfg compute probable parse user query reflecting linguistic structure word

#* results challenges web search evaluation
#@David Hawking,Nick Craswell,Paul B. Thistlewaite,Donna Harman
#t1999
#cComputer Networks
#index58
#id58
#%
#! frozen million page snapshot web created enable andencourage meaningful reproducible evaluation web search systems techniques collection used evaluation framework text retrieval ference trec hopefully provide convincing answers questions canlink information result better rankings longer queries result better answers trec systems work web data snapshot associated evaluationmethods described invitation extended participate preliminary results arepresented effectivess comparison trec systems working snapshot col lection known web search systems working current web thesesuggest standard document rankings produced public web search engines isby means state art

#* scalable browsing large collections case study
#@Gordon W. Paynter,Ian H. Witten,Sally Jo Cunningham,George Buchanan
#t2000
#cACM DL
#index59
#id59
#%
#! phrase browsing techniques use phrases extractedautomatically large information collection basisfor browsing accessing itfi paper describes casestudy uses automatically constructed phrasehierarchy facilitate browsing ordinary large websitefi phrases extracted text using novelcombination rudimentary syntactic processing andsequential grammar induction techniquesfi interface issimple robust easy usefito convey

#* searching web public queries
#@A Spink,D Wolfram,MBJ Jansen,T Saracevic
#t2001
#cJournal of the American Society for Information Science and Technology
#index60
#id60
#%
#! studying actual web searching public large analyzed million web queries users excite search engine people use search terms modified queries view web pages rarely use advanced search featuresfi small number search terms used high frequency great terms unique language web queries distinctivefi queries recreation en tertainment rank highestfi findings compared data large studies web queriesfi study provides insight public practices choices web searchingfi

#* index structures efficient phrase querying
#@Hugh E. Williams,Justin Zobel,Phil Anderson
#t1999
#cAustralasian Database Conference
#index61
#id61
#%
#! text retrieval systems used fetch documents large text collections using queries consisting words word sequences

#* managing gigabytes compressing indexing documents images
#@Ian H. Witten,Alistair Moffat,Timothy C. Bell
#t1994
#cnull
#index62
#id62
#%
#! fully updated second edition highly acclaimed managing gigabytes authors witten moffat bell continue provide unparalleled coverage state art techniques compressing indexing data field work large quantities information book essential reading authoritative theoretical resource practical guide meeting toughest storage access challenges covers latest developments compression indexing application web digital libraries details dozens powerful techniques supported mg authors compressing storing retrieving text images textual images mg source code freely available web

#* exploring similarity space
#@Justin Zobel,Alistair Moffat
#t1998
#cSIGIR Forum
#index63
#id63
#%
#! ranked queries used locate relevant documents text database ranked query list terms specified documents closely match query returned decreasing order similarity answersfi crucial efficacy ranked querying use similarity heuristic mechanism assigns numeric score indicating closely document query matchfi note explore categorise range similarity heuristics described

#* optimization inverted vector searches
#@Chris Buckley,A. F. Lewit
#t1985
#cSIGIR
#index64
#id64
#%
#! simple algorithm presented increasing efficiency information retrieval searches implemented using inverted files optimization algorithm employs knowledge methods used weighting document query terms order examine inverted lists possible extension basic algorithm allows greatly increased performance optimization modest cost retrieval effectiveness experimental runs examining different term weighting models showing optimization possible

#* new retrieval approaches using smart trec
#@Chris Buckley,Amit Singhal,Mandar Mitra
#t1995
#cTREC
#index65
#id65
#%
#! smart information retrieval project emphasizes completely automatic approaches understandingand retrieval large quantities textfi continue work trec performing runs therouting ad hoc confused text interactive foreign language environmentsfiintroductionfor years smart project cornell university interested analysis search andretrieval heterogeneous text databases vocabulary allowed vary widely

#* indexing latent semantic analysis
#@Scott C. Deerwester,Susan Dumais,Thomas K. Landauer,George W. Furnas,Richard A. Harshman
#t1990
#cJASIS
#index66
#id66
#%
#! new method automatic indexing retrieval described approach advantage implicit higher order structure association terms documents semantic structure order improve detection relevant documents basis terms queries particular technique used singular value decomposition large term document matrix decomposed set ca thogonal factors original matrix approximated linear combination documents represented ca item vectors factor weights queries represented pseudo document vectors formed weighted combinations terms documents supra threshold cosine values turned initial tests completely automatic method retrieval promising

#* retrieving records gigabyte text minicomputer using statistical ranking
#@Donna Harman, C Gerald
#t1990
#cJournal of the American Society for Information Science
#index67
#id67
#%
#! statistically based ranked retrieval records using keywords provides advantages traditional boolean retrieval methods especially end users approach retrieval seen widespread use large operational retrieval systems feasibility retrieval methodology research produce fast search techniques using ranking algorithms test results large databases end users results response times order seconds megabytes text favorable user reaction novice users able consistently obtain good search results minutes training additional work devise new indexing techniques create inverted files large databases using minicomputer techniques use sorting require working space size input text produce indices input text size

#* text indexing based lexical relations application software libraries
#@Y. S. Maarek,F.Z.Smadja
#t1989
#cACM SIGIR Forum
#index68
#id68
#%
#! contrast kinds libraries software libraries need conceptually organized looking component main concern users functionality desired component implementation details secondary software reuse enhanced conceptually organized large libraries software components paper present guru tool allows automatical building large software libraries documented software components focus guru indexing component extracts conceptual attributes natural language documentation indexing method based words occurrences uses extract occurrence knowledge compiler extracting potential attributes textual documents conceptually relevant collocations selected according resolving power scales noise context words fully automated indexing tool goes keyword based tools understanding document brittleness knowledge based tools indexing component guru fully implemented results given paper

#* fast ranking limited space
#@Alistair Moffat,Justin Zobel
#t1994
#cICDE
#index69
#id69
#%
#! ranking techniques long suggested alternatives conventionalboolean methods searching document collectionsfi costof computing ranking greater cost performing aboolean search terms memory space processing timefi herewe consider resources required cosine method ranking andshow careful application indexing selection techniquesboth space time required ranking substantially

#* overview seventh text retrieval conference trec
#@EM Voorhees,Donna Harman
#t1998
#cProceedings of the Seventh Text REtrieval Conference (TREC-7)
#index70
#id70
#%
#! paper serves introduction research described remainder volumefi itconcentrates mainly main task ad hoc retrieval defined sectionfi details regardingthe test collections evaluation methodology used trec follow sections section provides overview ad hoc retrieval resultsfi addition main ad hoc task trec containedseven tracks tasks focus research particular subproblems

#* data compression fulltext retrieval systems
#@Timothy C. Bell, Alistair Moffat, Craig G. Nevill-Manning,Ian H. Witten,Justin Zobel
#t1993
#cJournal of the American Society for Information Science archive
#index71
#id71
#%
#! describes compression methods components text systems text databases cd rom topics discussed include storage media structures text retrieval including indexes inverted files bitmaps compression tools memory requirements retrieval ranking information retrieval

#* model based concordance compression
#@Abraham Bookstein,Shmuel T. Klein,Timo Raita
#t1992
#cData Compression Conference
#index72
#id72
#%
#! authors discuss concordance compression using framework customary compression theory begin creating mathematical model concordance generation use optimal compression engines huffman arithmetic coding actual compression noted context static information retrieval compression decompression symmetrical tasks compression building decompression needed processing query directly affects response time use extensive costly preprocessing compression provided reasonably fast decompression methods possible compression applied files text concordance decompression needed possibly short pieces accessed random means pointers exact locations use adaptive methods based tables systematically change beginning end file ruled concern speed encoding decoding relating concordance compression conceptually modern approach data compression testing effectiveness models

#* fundamentals database systems
#@Ramez Elmasri,Shamkant B. Navathe
#t1989
#cnull
#index73
#id73
#%
#! clear explanations theory design broad coverage models real systems date introduction modern database technologies result leading introduction database systems fresh new problems new lab manual students opportunities practice fundamentals design implementation real world examples serve engaging practical illustrations database concepts fifth edition maintains coverage popular database topics including sql security data mining contains new chapter web script programming databases

#* access methods text
#@Christos Faloutsos
#t1985
#cACM Comput. Surv.
#index74
#id74
#%
#! paper compares text retrieval methods intended office systems operational requirements office environment discussed retrieval methods database systems information retrieval systems examined classify methods examine interesting representatives class attempts speed retrieval special purpose hardware presented issues approximate string matching compression discussed qualitative comparison examined methods presented signature file method discussed

#* signature files access method documents analytical performance evaluation
#@Christos Faloutsos,Stavros Christodoulakis
#t1984
#cACM Trans. Inf. Syst.
#index75
#id75
#%
#! signature file access method text retrieval studied according method documents stored sequentially text file ions signatures documents stored signature file serves filter retrieval helps discarding large number nonqualifying documents paper methods creating signatures studied analytically based word signatures superimposed coding closed form formulas derived false drop probability methods factors affect studied performance comparisons methods based formulas provided

#* optimal source codes geometrically distributed integer alphabets
#@R Gallager,D Van Voorhis
#t1975
#cInformation Theory, IEEE Transactions on
#index76
#id76
#%
#! letp theta theta ibe probability assignment set nonnegative integers wherethetais arbitrary real number theta optimal binary source code probability assignment constructed follows letlbe integer satisfyingtheta theta leq theta theta represent nonnegative integeriasi lj rwhenj lfloor rfloor integer ofi andr mod encodejby unary code jzeros followed single encoderby huffman code using codewords lengthlfloor log rfloor forr lfloor log rfloor lengthlfloor log rfloor optimal code nonnegative integers concatenation codes

#* run length encodings
#@
#t1000
#cEMPTY
#index77
#id77
#%
#! article discuss implement simple compression algorithm rle fact easy good introduction programmers interested data compression rle mainly used compress runs byte rle good stage bwt compressor ll avoid time sorting strings equal fact lot runs bytes data

#* storing text retrieval systems cd rom compression encryption considerations
#@Shmuel T. Klein,Abraham Bookstein,Scott C. Deerwester
#t1989
#cSIGIR
#index78
#id78
#%
#! emergence cd rom storage medium text databases raises question maximum size database contained medium example problem storing tr sor la langue fran ccidel aise cd rom examined paper text database megabytes long cd rom hold addition dictionary concordance needed access data stored constraint material copyrighted desirable material difficult decode software provided pertinent approaches compression various files reviewed compression text related problem data encryption specifically shown simple models text generation huffman encoding produces bit string indistinguishable representation coin flips

#* compression indexes positional information large text databases
#@Gordon Linoff,Craig Stanfill
#t1993
#cSIGIR
#index79
#id79
#%
#! paper describes combination compression methods used reduce size inverted indexes large text databases methods prefix omission run length encoding novel family numeric representations called coding using compression methods different text sources king james version bible sample wall street journal stories compressed index occupies size original text stopwords numbers included index decreased time required fully compensate time needed uncompress postings research effort handle large text databases cm massively parallel mimd supercomputer

#* development stemming algorithm
#@Julie B. Lovins
#t1968
#cMechanical Translation and Computational Linguistics
#index80
#id80
#%
#! stemming algorithm procedure reduce words stem common form useful areas computational linguistics information retrieval work form algorithm varies applications certain linguistic problems common stemming procedure basis evaluation previous attempts deal problems paper discusses theoretical practical attributes stemming algorithms new version context sensitive longest match stemming algorithm english developed use library information transfer general applications proposed major linguistic problem stemming variation spelling stems discussed feasible programmed solutions outlined sample results methods

#* inverted index implementation
#@Ken J. McDonell
#t1977
#cComput. J.
#index81
#id81
#%
#! working implementation inverted index suitable real time applications described based hash addressed random access organization variable length records index structure processing algorithms developed aid simulation model typical simulation results presented critical parameters identified solution inverted key identification problem proposed severe time penalty accrues designing software directly dependent main file record format

#* parameterised compression sparse bitmaps
#@Alistair Moffat,Justin Zobel
#t1992
#cSIGIR
#index82
#id82
#%
#! text retrieval systems typically use bitmap inverted fileto identify documents contain words documents containing anycombination words quickly locatedfi bitmaps word occurrences large usually sparse amenable variety compression techniquesfihere consider techniques encoding bitvector bitmap isparameterised different code used bitvectorfi

#* memory efficient ranking
#@Alistair Moffat,Justin Zobel,Ron Sacks-Davis
#t1994
#cInf. Process. Manage.
#index83
#id83
#%
#! fast effective ranking collection documents respect query requires structures including vocabulary inverted file entries arrays term weights document lengths array partial similarity accumulators address tables inverted file entries documentsfi structures array document lengths array accumulators components accessed frequently ranked query crucial acceptable

#* multi key access methods based superimposed coding techniques
#@R. Sacks-Davis,A. Kent,K. Ramamohanarao
#t1987
#c1987	ACM Transactions on Database Systems
#index84
#id84
#%
#! single level level indexed descriptor schemes multikey retrieval presented compared descriptors formed using superimposed coding techniques stored using bit inversion technique fast batch insertion algorithm cost forming bit inverted file disk access record presented large data files shown level implementation generally efficient queries small number matching records queries specify values potential problem level implementation costs accrue blocks records match query individual records blocks approach overcoming problem set bits descriptors based pairs indexed terms approach presented analyzed

#* introduction modern information retrieval
#@Gerard Salton,Michael McGill
#t1984
#cnull
#index85
#id85
#%
#! new technology allows design sophisticated information retrieval analyze process store retrieve specific resources matching particular user needs clear practical text relates theory techniques tools critical making information retrieval work completely revised second edition incorporates latest developments rapidly expanding field including multimedia information retrieval user interfaces digital libraries chowdhury coverage comprehensive including classification cataloging subject indexing ing vocabulary control cd rom online information retrieval multimedia hypertext hypermedia expert systems natural language processing user interface systems internet world wide web digital library environments illustrated examples comprehensively referenced international audience ideal textbook students library information studies professionals eager advance knowledge future information

#* efficient indexing technique text database systems
#@Justin Zobel,Alistair Moffat,Ron Sacks-Davis
#t1992
#cProceedings of 18th International Conference on Very Large Databases
#index86
#id86
#%
#! text database systems require index allow fast access documents based content propose inverted file indexing scheme based compression scheme allows users retrieve documents using words occurring documents sequences adjacent words statistical ranking techniques compression methods chosen ensure storage requirements small dynamic update straightforward assumption make sufficient main memory available support memory vocabulary given assumption method requires disc access query term identify answers queries

#* data compression text retrieval systems
#@Timothy C. Bell,Alistair Moffat,Craig G. Nevill-Manning,Ian H. Witten,Justin Zobel
#t1993
#cJASIS
#index87
#id87
#%
#! describes compression methods components text systems text databases cd rom topics discussed include storage media structures text retrieval including indexes inverted files bitmaps compression tools memory requirements retrieval ranking information retrieval

#* optimisation inverted vector searches
#@Chris Buckley,Alan F. Lewit
#t1985
#cAnnual ACM Conference on Research and Development in Information Retrieval
#index88
#id88
#%
#! review use inverted files best match searching information retrieval systems

#* coding compression text retrieval systems
#@Alistair Moffat,Justin Zobel
#t1992
#cData Compression Conference
#index89
#id89
#%
#! witten bell nevill ibid described compression models use text retrieval systems authors discuss coding methods use models results scheme yielding virtually identical compression decoding times faster main features implementation complete absence arithmetic coding reason high speed implementation particularly suited slow devices cd rom answering query requires disk access term query disk access answer words numbers indexed stop words built compressed databases

#* review use inverted files best match searching information retrieval systems
#@Shirley A. Perry,Peter Willett
#t1983
#cJournal of Information Science
#index90
#id90
#%
#! use inverted files calculation similarity coefficients types matching function discussed context mechanised document retrieval systems critical evaluation presented range algorithms described matching documents queries particular attention paid computational efficiency various procedures improved search heuristics given cases suggested algorithms implemented sufficiently efficiently permit provision nearest neighbour searching standard retrieval option

#* fast high quality multilevel scheme partitioning irregular graphs
#@G Karypis,Vipin Kumar
#t1998
#cSIAM Journal on Scientific Computing
#index91
#id91
#%92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122
#! recently number researchers investigated class graph partitioning algorithms reduce size graph collapsing vertices edges partition smaller graph uncoarsen construct partition original graph early work clear multilevel techniques held great promise known consistently produce high quality partitions graphs arising wide range application domains investigate ectiveness di erent choices phases coarsening partition coarsest graph refinement particular present new coarsening heuristic called heavy edge heuristic size partition coarse graph small factor size final partition obtained multilevel refinement present faster variation kernighan lin kl algorithm refining uncoarsening test scheme large number graphs arising various domains including finite element methods linear programming vlsi transportation experiments scheme produces partitions consistently better produced spectral partitioning schemes substantially smaller time scheme used compute reducing orderings sparse matrices produces orderings substantially smaller widely used multiple minimum degree algorithm

#* fast multilevel implementation recursive spectral bisection partitioning unstructured problems
#@Stephen T. Barnard,Horst D. Simon
#t1993
#cPPSC
#index92
#id92
#%
#! unstructured meshes used large scale scientific engineering problems including finite volume methods computational fluid dynamics finite element methods structural analysis unstructured problems solved distributed memory parallel computers data structures partitioned distributed processors solved efficiently partitioning maximize load balance minimize interprocessor

#* algorithm partitioning nodes graph
#@Earl R. Barnes
#t1981
#cDecision and Control including the Symposium on Adaptive Processes, 1981 20th IEEE Conference on
#index93
#id93
#%
#! let undirected graph having nodes edges consider problem partitioning disjoint subsets cdots given sizes cdots respectively way number edges connect different subsets minimal obtain heuristic solution solution linear programming transportation problem

#* graph bisection algorithms good average case behavior
#@Thang Nguyen Bui,Soma Chaudhuri,Frank Thomson Leighton,Michael Sipser
#t1987
#cCombinatorica
#index94
#id94
#%
#! paper polynomial time algorithm input graph outputs minimum bisection graph halts output importantly algorithm chooses course high probability natural classes graphs particular fixedd sufficiently largen allb algorithm finds minimum bisection alld regular labelled simple graphs nodes bisection widthb example algorithm succeeds regular graphs nodes bisection widtho algorithm differs graph bisection heuristics heuristics np complete problems respects

#* geometric spectral partitioning
#@Tony F. Chan,John R. Gilbert,Shang-Hua Teng
#t1995
#cnull
#index95
#id95
#%
#! investigate new method partitioning graph equal sized pieces connecting edgesfiwe combine ideas recently suggested partitioning algorithms spectral bisection uses eigenvectorof matrix associated graph geometric bisection applies graphs meshesin euclidean space fi new method does require geometric coordinates produces partitions areoften better spectral geometric onesfi

#* parallel algorithms dynamically partitioning unstructured grids
#@Pedro C. Diniz,Steve Plimpton,Bruce Hendrickson,Robert W. Leland
#t1995
#cPPSC
#index96
#id96
#%
#! grid partitioning method choice decomposing wide variety computationalproblems naturally parallel piecesfi problems computational loadon grid grid itself changes simulation progresses ability repartitiondynamically parallel attractive achieving higher performancefi wedescribe algorithms suitable parallel dynamic load balancing attemptto partition unstructured grids computational load

#* linear time heuristic improving network partitions
#@C. M. Fiduccia,R. M. Mattheyses
#t1982
#cAnnual ACM IEEE Design Automation Conference
#index97
#id97
#%
#! fiduccia matteyses min cut heuristic provides efficient solution theproblem separating network vertices separate partitions effort tominimize number nets contain nodes partition heuristic isdesigned specifically handle large complex networks contain multi terminaland weighted cells worst case computation time heuristic increaseslinearly overall size network additionally practice seen thatthe number iterations typically required cutest converge theminimum value typically small key factor obtaining linear timebehavior fact algorithm moves node time partitionsin attempt reduce current cut set maximum possible value alsonote times maximum cut set reduction negative pointwe proceed algorithm allows opportunity escape local minima currently movement node appropriate nodesconnected moved cell updated represent use simple yetefficient data structures allows ability avoid redundant searching best cell tobe moved similarly prevents excess unnecessary updates neighborcells need updated data structures assist adding thealready efficient nature algorithm heuristic contains balance constraintwhich allows user retain control size created partitions

#* finding clusters vlsi circuits
#@Jorn Garbers,Hans Jurgen Promel,Angelika Steger
#t1990
#cICCAD
#index98
#id98
#%
#! circuit partitioning plays fundamental role hierarchical layout systems identifying strongly connected subcircuits clusters logic significantly reduce delay circuit total interconnection length finding cluster partition np complete authors propose fast heuristic algorithm based simple local criterion able prove highly structured circuits clusters algorithm correspond high probability natural clusters application large scale real world circuits shows method number nets cut reduced compared standard mincut approach

#* evolution minimum degree ordering algorithm
#@Alan George,Joseph W.H. Liu 1989
#t0
#cnull
#index99
#id99
#%
#! past years implementation minimum degree algorithm received study important enhancements article various enhancements trace historical development provide experiments showing effective improving execution time algorithm present shortcoming exists widely used implementations algorithm quality ordering provided implementations surprisingly sensitive initial ordering example changing input ordering lead increase decrease factor cost subsequent numerical factorization sensitivity caused lack effective tie breaking strategy experiments illustrate importance developing strategy

#* geometric mesh partitioning implementation experiments
#@John R. Gilbert,Gary L. Miller,Shang-Hua Teng
#t1995
#cIPPS
#index100
#id100
#%
#! investigate method dividing irregular mesh equal sized pieces interconnecting edgesfi method novel feature exploits geometric coordinates mesh verticesfi based theoretical work miller teng thurston vavasis showed certain classes shaped finite element meshes good separatorsfi geometric method quite simple implement matlab code detailfi method quite

#* heuristic algorithms automatic graph partitioning
#@T. Goehring,Y. Saad
#t1995
#cTechnical ReportUMSI
#index101
#id101
#%
#! practical implementations finite element method distributed memory multi systems necessitate use partitioning tools subdivide mesh sub meshes roughly equal size graph partitioning algorithms mandatory implementing distributed sparse matrix methods domain decomposition techniques irregularly structured problems parallel computers propose class algorithms based level set expansions number center nodes

#* highly scalable parallel algorithms sparse matrix factorization
#@Anshul Gupta,G Karypis,Vipin Kumar
#t1997
#cIEEE Trans. Parallel Distrib. Syst.
#index102
#id102
#%
#! paper scalable parallel algorithms sparse matrix factorization analyze performance scalability present experimental results processors cray parallel analysis experimental results demonstrate algorithms substantially improve state art parallel direct solution sparse linear systems terms scalability overall performance known fact dense matrix factorization scales implemented efficiently parallel computers paper present algorithms factor wide class sparse matrices including arising dimensional finite element problems asymptotically scalable dense matrix factorization algorithms variety parallel architectures algorithms incur communication overhead scalable previously known parallel formulation sparse matrix factorization paper discuss cholesky factorization symmetric positive definite matrices algorithms adapted solving sparse linear squares problems gaussian elimination diagonally dominant matrices symmetric structure implementation sparse cholesky factorization algorithms delivers gflops cray medium size structural engineering linear programming problems best knowledge highest performance obtained sparse cholesky factorization supercomputer

#* fast spectral methods ratio cut partitioning clustering
#@Lars W. Hagen,Andrew B. Kahng
#t1991
#cICCAD
#index103
#id103
#%
#! ratio cut partitioning objective function successfully embodies traditional min cut equipartition goals partitioning fiduccia mattheyses style ratio cut heuristics achieved cost savings averaging circuit partitioning hardware simulation applications authors theoretical correspondence optimal ratio cut partition cost second smallest eigenvalue particular netlist derived matrix present fast lanczos based methods computing heuristic ratio cuts eigenvector second eigenvalue results better previous methods average primary mcnc benchmarks efficient clustering method based second eigenvector successful difficult input classes cad aided design literature extensions directions future work considered

#* new approach ective circuit clustering
#@Lars Hagen,Andrew B. Kahng
#t1992
#cIEEE International Conference on Computer Aided Design
#index104
#id104
#%
#! complexity generation vlsi systems exceed capabilities layout synthesis algorithms particularly netlist partitioning module placement clustering needed condense netlist problem size tractable existing optimization methods paper establish ds qua lity measure general metric evaluation clustering algorithms ds metric turn motivates rwst algorithm new self tuning clustering method based random walks circuit netlist rwst efficiently captures globally good circuit clustering incorporated phase iterative fiduccia mattheyses partitioning strategy rw st clustering method improves bisection width average previous maiching based methods

#* parallel algorithms matrix computations
#@K. A. Gallivan,Michael T. Heath,Esmond Ng,James M. Ortega,Barry W. Peyton,R. J. Plemmons,Charles H. Romine,A. H. Sameh,Robert G. Voigt
#t1990
#cbook
#index105
#id105
#%
#! shown mesh connected multiprocessor compute inverse matrix linear time algorithm based theorem known sylvester computes cofactor matrix steps involves unit distance message routing arithmetic operations processor coding memory requirement processor independent shown algorithm solves systems linear equations linear time processors

#* cartesian parallel nested dissection algorithm
#@Michael T. Heath,Padma Raghavan
#t1994
#cSIAM Journal on Matrix Analysis and Applications
#index106
#id106
#%
#! paper concerned distri uted parallel computation ordering symmetric positive nite sparse matrix purpose ordering limit ll enhance concurrency su se uent computation cholesky factori ation matrix use geometric approach nested dissection ased given cartesian em edding graph matrix euclidean space resulting algorithm implemented ciently massively parallel distri uted memory

#* improved spectral graph partitioning algorithm mapping parallel computations
#@Bruce Hendrickson,Robert Leland
#t1995
#cSIAM Journal on Scientific Computing
#index107
#id107
#%
#! efficient use distributed memory parallel requires computational load balanced processors way minimizes interprocessor communication new domain mapping algorithm presented extends recent work ideas spectral graph theory applied problem generalization spectral graph bisection involves novel use multiple eigenvectors allow division computation parts stage recursive decomposition resulting method suitable scientific computations like irregular finite elements differences performed hypercube mesh architecture machines experimental results confirm new method provides better decompositions arrived economically robustly previous spectral methods algorithm allows arbitrary nonnegative weights vertices edges model inhomogeneous computation communication new spectral lower bound graph bisection presented

#* multilevel algorithm partitioning graphs
#@Bruce Hendrickson,Robert Leland
#t1995
#cConference on High Performance Networking and Computing
#index108
#id108
#%
#! graph partitioning problem dividing vertices graph sets specified sizes edges cross setsfi np complete problem arises important scientific engineering problemsfi prominent examples include mapping parallel computations laying circuits ordering sparse matrix computationsfi present multilevel algorithm graph partitioning graph approximated sequence increasingly

#* parallel formulation interior point algorithms
#@G Karypis,Anshul Gupta,Vipin Kumar
#t1994
#cSC
#index109
#id109
#%
#! recent years interior point algorithms used successfully solving medium large size linear programming lp problems paper highly parallel formulation interior point algorithm key component interior point algorithm solution sparse linear equations using cholesky factorization performance parallel cholesky factorization determined communication overhead incurred algorithm load imbalance processors parallel interior point algorithm use recently developed parallel multifrontal algorithm smallest communication overhead parallel algorithms cholesky factorization developed date computation imbalance depends shape elimination tree associated sparse reordered factorization balance computation implemented evaluated di erent ordering algorithms algorithms kernighan lin spectral nested dissection yield balanced elimination trees greatly increase parallelism exploited preliminary implementation achieves speedup high processor ncube moderate size problems

#* analysis multilevel graph partitioning
#@G Karypis,Vipin Kumar
#t1995
#cSC
#index110
#id110
#%
#! recently number researchers investigateda class algorithms based multilevel graphpartitioning moderate computational complexity provide excellent graph partitionsfi exist little theoretical analysis explain abilityof multilevel algorithms produce good partitionsfi thispaper present analysisfi certainreasonable assumptions refinement usedin uncoarsening phase

#* parallel algorithm multilevel graph partitioning sparse matrix ordering
#@G Karypis,Vipin Kumar
#t1998
#cJ. Parallel Distrib. Comput.
#index111
#id111
#%
#! paper present parallel formulation multilevel graph partitioning sparse matrix orderingalgorithmfi key feature parallel formulation distinguishes proposed parallel formulations ofmultilevel algorithms partitions vertices graph parts distributing overall adjacencymatrix graph processorsfi mapping results substantially smaller communication onedimensionaldistribution

#* introduction parallel computing design analysis algorithms
#@
#t1000
#cEMPTY
#index112
#id112
#%
#! introduction parallel computing provides depth look techniques design analysis parallel algorithms programming algorithms commercially available parallel platforms book discusses principles parallel algorithm design different parallel programming models extensive coverage mpi posix threads openmp provides broad balanced coverage various core topics sorting graph algorithms discrete optimization techniques data mining algorithms number algorithms used numerical scientific computing applications basic approach advocated text portable parallel algorithm software development empahsis lacking existing textbooks parallel computing enhance pedagogical value text extensive examples diagrams exercises varying degrees difficulty bibliographical remarks provided addition serving textbook reference source professionals parallel software developers book help students researchers non science disciplines need solve computation intensive problems using parallel computers

#* separator theorem planar graphs
#@Richard J. Lipton,Robert E. Tarjan
#t1979
#cSIAM Journal on Applied Mathematics
#index113
#id113
#%
#! let vertex planar graph prove verticesof partitioned sets edgejoins vertex vertex containsmore vertices contains vertices exhibit algorithm finds partition time

#* modification minimum degree algorithm multiple elimination
#@Joseph W. H.
#t1985
#cACM Trans. Math. Softw
#index114
#id114
#%
#! widely used ordering scheme reduce fills operations sparse matrix computation minimum degree algorithm notion multiple elimination introduced modification conventional scheme motivation discussed using grid model problem experimental results indicate modified version retains reducing property better original ordering algorithm requires time reduction ordering time problem dependent problems modified algorithm run times faster existing implementations minimum degree algorithm use external degree algorithm introduced

#* automatic mesh partitioning
#@Gary L. Miller,Shang Teng,William Thurston,Stephen A. Vavasis
#t1992
#cTechnical Report
#index115
#id115
#%
#! paper describes efficient approach partitioning unstructured meshes occur naturally finite element finite difference methods approach makes use underlying geometric structure given mesh finds provably good partition random time applies meshes dimensions new method applications efficient sequential parallel algorithms large scale problems scientific computing overview paper written emphasis algorithmic aspects approach detailed proofs companion papers

#* unified geometric approach graph separators
#@Gary L. Miller,Shang-Hua Teng,Stephen A. Vavasis
#t1991
#cFOCS
#index116
#id116
#%
#! class graphs called overlap graphs proposed special cases overlap graphs include planar graphs nearest neighbor graphs earlier classes graphs associated finite element methods separator bound proved overlap graphs embedded dimensions result unifies earlier separator results arguments based geometric properties embedding separator bounds come randomized linear time randomized nc algorithms bounds best possible leading term

#* validity oriented approach partitioning large sparse graphs connectivity constraint
#@P Ciarlet,F Lamour
#t1996
#cNumerical Algorithms
#index117
#id117
#%
#! paper consider problem partitioning large sparse graphs finite element meshes heuristic proposed allows partition connected quasi balanced subgraphs reasonable time attempting minimize number edge cuts goal build partitions graphs containing large numbers nodes edges practice basically algorithm relies iterative construction connected subgraphs construction achieved successively exploring clusters nodes called fronts judicious use fronts ensures connectivity subsets low cost shown locally given subgraph complexity operations grows linearly number edges examples given illustrate quality speed heuristic

#* combinatorial optimization
#@Vijay Chandru,M. R. Rao
#t1997
#cThe Computer Science and Engineering Handbook
#index118
#id118
#%
#! comprehensive textbook combinatorial optimization puts special emphasis theoretical results algorithms provably good performance contrast heuristics arisen basis courses combinatorial optimization special topics graduate level complete book contains material semesters hours week usually selects material suitable way book contains complete concise proofs deep results did appear book recent topics covered references provided book represents state art combinatorial optimization edition contains new chapter facility location problems area extremely active past years furthermore new sections material various topics new exercises updates bibliography added reviews nd edition book combinatorial optimization beautiful example ideal textbook operations resarch letters second edition corrections updates recommendable book documents relevant knowledge combinatorial optimization records problems algorithms define discipline today read stimulating researchers practitioners students interested combinatorial optimization

#* computing block triangular form sparse matrix
#@Alex Pothen,Chin-Ju Fan
#t1990
#cACM Trans. Math. Softw.
#index119
#id119
#%
#! consider problem permuting rows columns rectangular square unsymmetric sparse matrix compute block triangular form block triangular form based canonical decomposition bipartite graphs induced maximum matching discovered dulmage mendelsohn implementations algorithms compute block triangular form provide computational results sparse matrices test collections applications block triangular form included

#* partitioning sparse matrices eigenvectors graphs
#@Alex Pothen,Horst D. Simon,Kan-Pu Liou
#t1990
#cSIAM Journal on Matrix Analysis and Applications
#index120
#id120
#%
#! problem computing small vertex separator graph arises context computing good ordering parallel factorization sparse symmetric matrices algebraic approach computing vertex separators considered paper shown lower bounds separator sizes obtained terms eigenvalues laplacian matrix associated graph laplacian eigenvectors grid graphs computed kronecker products involving eigenvectors path graphs eigenvectors used compute good separators grid graphs heuristic algorithm designed compute vertex separator general graph computing edge separator graph eigenvector laplaeian matrix using maximum matching subgraph compute vertex separator results quality separators computed spectral algorithm presented compared separators obtained automatic nested dissection kernighan lin algorithm finally report time required compute laplacian eigenvector consider accuracy eigenvector computed obtain good separators spectral algorithm advantage implemented medium size multiprocessor straight forward manner

#* spectral nested dissection
#@Alex Pothen,Horst D. Simon,Lie Wang
#t1992
#c1992	Technical Report
#index121
#id121
#%
#! fi spectral nested dissection algorithm computing orderings appropriate forparallel factorization sparse symmetric matricesfi algorithm makes use spectral properties thelaplacian matrix associated given matrix compute separatorsfi evaluate quality spectral orderings respect measures elimination tree height height weight balancesof elimination trees clique tree heightsfi spectral orderings compare quite

#* fast implementation spectral nested dissection
#@Alex Pothen,Horst D. Simon,Lie Wang,Stephen T. Barnard
#t1992
#cSC
#index122
#id122
#%
#! authors novel spectral nested dissection snd algorithm novel algorithm computing orderings appropriate parallel factorization sparse symmetric matrices algorithm makes use spectral properties laplacian matrix associated given matrix compute separators authors evaluate quality spectral orderings respect measures elimination tree height height weight balances elimination trees clique tree heights use large structural analysis problems test cases demonstrate real applications spectral orderings compare quite favorably commonly used orderings outperforming wide margin measures disadvantage snd relatively long execution time

#* highly adaptive distributed routing algorithm mobile wireless networks
#@Vincent D. Park,M Corson
#t1997
#cINFOCOM
#index123
#id123
#%124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139
#! present new distributed routing protocol mobile multihop wireless networks protocol family protocols term link reversal algorithms protocol reaction structured temporally ordered sequence diffusing computations computation consisting sequence directed link reversals protocol highly adaptive efficient scalable best suited use large dense mobile networks networks protocol reaction link failures typically involves localized single pass distributed algorithm capability unique protocols stable face network partitions results protocol high degree adaptivity desirable behavior achieved novel use physical logical clock establish temporal order topological change events used structure order algorithm reaction topological changes refer protocol temporally ordered routing algorithm tora

#* architectural considerations mobile mesh networking
#@M Corson,J Macker,S Batsell
#t1996
#cMILCOM
#index124
#id124
#%
#! mobile communications central military operations necessary communicate simultaneously multiple warfighters engaged common task paper describes problem routing resource reservation mobile mesh networks presents architectural recommendations necessary internet protocols operate effectively environments mobile mesh network autonomous mobile routers connected wireless links union form arbitrary graph routers free randomly network wireless topology change rapidly unpredictably

#* failsafe distributed routing protocol
#@P Merlin,A Segall
#t1979
#c1979	Communications, IEEE Transactions on
#index125
#id125
#%
#! algorithm constructing adaptively maintaining routing tables communication networks presented algorithm employed message circuit switching networks uses distributed computation provides routing tables loop free destination times adapts changes network flows completely failsafe means arbitrary failures additions network recovers finite time sense providing routing paths physically connected nodes destination routes independently updated update cycle triggered destination

#* responsive distributed routing algorithm networks
#@Jeffrey M. Jaffe,Franklin H. Moss
#t1981
#cICDCS
#index126
#id126
#%
#! new distributed algorithm presented dynamically determining weighted shortest paths used message routing networks major features algorithm paths defined form transient loops weights change number steps required new shortest paths network links fail previous algorithms specifically worst case recovery time proportional largest number hopshin weighted shortest paths previous loop free distributed algorithms recovery time proportional

#* adaptive shortest path algorithm
#@
#t1000
#cEMPTY
#index127
#id127
#%
#! authors distributed algorithm compute shortest paths network changing topology authors analyze behavior proof correctness discussed does suffer routing table looping behavior associated ford bellman distributed shortest path algorithm uses truly distributed processing time message complexities evaluated comparisons methods given

#* distributed routing labeled distances
#@J.J. Garcia-Luna-Aceves
#t1992
#cINFOCOM
#index128
#id128
#%
#! author presents verifies analyzes new routing algorithm called labeled distance vector routing algorithm ldr loop free instant eliminates counting infinity problem distributed bellman ford dbf algorithm operates arbitrary link node delays provides shortest paths finite time occurrence arbitrary sequence topological changes contrast previous successful approaches loop free routing ldr maintains dbf row independence property does require internodal coordination spanning multiple loops new algorithm shown loop free converge finite time arbitrary sequence topological changes performance compared performance distributed routing algorithms

#* loop free routing using diffusing computations
#@J.J. Garcia-Luna-Aceves
#t1993
#cIEEE/ACM Trans. Netw.
#index129
#id129
#%
#! family distributed algorithms dynamic computation shortest paths network memet presented validated analyzed according algorithms node maintains vector distance node update messages node sent neighbors message contains dktance vector entries entry specifies length selected path network destination indication entry constitutes update query reply previous query new algorithms treat problem distributed shortest path routing diffusing computations firzt proposed dijkztra scholten improve algorithms introduced previously chandy misra jatye moss merlin segatl author new algorithms shown converge finite time arbitrary sequence link coat topological changes loop free instan outperform loop free routing algorithms previously proposed standpoin

#* loop free path finding algorithm specification verification complexity
#@J.J. Garcia-Luna-Aceves,S. Murthy
#t1995
#cProc. IEEE INFOCOM 95
#index130
#id130
#%
#! loopfree path finding algorithm lpa presented lpa specifies second hop distance destination ensure termination addition uses inter neighbor synchronization mechanism eliminate temporary loops detailed proof lpa correctness presented complexity evaluated lpa average performance compared simulation performance algorithnis representative state art distributed routing ideal link state ils algorithm loopfree algorithm based internodal coordination spanning multiple hops dual simulation results lpa scalable alternative dual ils terms average number steps messages operations needed algorithm converge topology change lpa shown achieve loop freedom instant additional overhead incurred prior algorithms based second hop distance information

#* distributed scalable routing based vectors link states
#@J.J. Garcia-Luna-Aceves,Jochen Behrens
#t1995
#cIEEE Journal on Selected Areas in Communications
#index131
#id131
#%
#! link vector algorithms lva introducedfor distributed maintenance routing information inlarge networks internetsfi according lva eachrouter maintains subset topology correspondsto adjacent links links used neighbor routersin preferred paths known destinationsfi based onthat subset topology information router derives itsown preferred paths communicates correspondinglink state information neighborsfi

#* distributed algorithms generating loop free routes networks frequently changing topology
#@MG ELI,P DIMITRI
#t1981
#cCommunications, IEEE Transactions on
#index132
#id132
#%
#! consider problem maintaining communication nodes data network central station presence frequent topological changes example mobile packet radio networks argue flooding schemes significant drawbacks networks propose ag eneral class distributed algorithms establishing new loop free routes station node left route china nthge sn etwork topology virtue built redundancy algorithms typically activatedv ery infrequently evenw hen involve communication portoiof nth network heen materially affected topological change

#* distributed routing algorithm mobile wireless networks
#@M Corson,A. Ephremides
#t1995
#cWireless Networks
#index133
#id133
#%
#! present loop free distributed routing protocol mobile packet radio networks protocol intended use networks rate topological change fast make flooding possible routing method slow make existing protocols nearly static topology applicable routing algorithm adapts asynchronously distributed fashion arbitrary changes topology absence global topological knowledge protocol uniqueness stems ability maintain source initiated loop free multipath routing desired destinations minimal overhead randomly varying topology protocol performance measured terms end end packet delay throughput compared pure flooding alternative algorithm suited high rate topological change environment envisioned protocol emphasis placed examining performance measures vary function rate topological changes network topology message traffic level results indicate new protocol generally outperforms alternative protocol rates change heavy traffic conditions opposite true light traffic protocols significantly outperform flooding rates change ultra high rates algorithms collapse network topology dense sparsely connected seen major factor relative performance algorithms

#* highly dynamic destinationsequenced distance vector routing dsdv mobile computers
#@Charles Perkins,Pravin Bhagwat
#t1994
#cACM SIGCOMM
#index134
#id134
#%
#! ad hoc network cooperative engagement collection mobile hosts required intervention centralized access point paper present innovative design operation ad hoc networks basic idea design operate mobile host specialized router periodically advertises view interconnection topology mobile hosts network amounts new sort routing protocol

#* efficient routing protocol wireless networks
#@S. Murthy,J.J. Garcia-Luna-Aceves
#t1996
#cMobile Networks and Applications
#index135
#id135
#%
#! present wireless routing protocol wrp wrp routing nodes communicate distance second hop destination wrp reduces number cases temporary routing loop occur accounts fast convergence properties detailed proof correctness presented performance compared simulation performance distributed bellman ford algorithm dbf dual loop free distance vector algorithm ideal link state algorithm ils represent state art internet routing simulation results indicate wrp efficient alternatives analyzed

#* dynamic source routing ad hoc wireless networks
#@David B. Johnson,David A. Maltz
#t1996
#cMobile Computing
#index136
#id136
#%
#! ad hoc network collection wireless mobile hosts forming temporary network theaid established infrastructure centralized administration environment benecessary mobile host enlist aid hosts forwarding packet destination limited range mobile host wireless transmissions paper presents protocolfor routing ad hoc networks uses dynamic source routing protocol adapts quickly routingchanges host movement frequent requires little overhead periods whichhosts frequently based results packet level simulation mobile hosts operating inan ad hoc network protocol performs variety environmental conditions hostdensity movement rates highest rates host movement simulated overhead ofthe protocol quite low falling just total data packets transmitted moderate movement ratesin network mobile hosts cases difference length routes used theoptimal route lengths negligible cases route lengths average factor optimal

#* network time protocol
#@DL Mills
#t1992
#cprotocol
#index137
#id137
#%
#! network time protocol ntp protocol synchronizing clocks systems packet switched variable latency data networks ntp uses udp port transport layer designed particularly resist effects variable latency jitter buffer

#* time clocks ordering events distributed
#@Leslie Lamport
#t1978
#cCommun. ACM
#index138
#id138
#%
#! concept event happening distributed examined shown define partial ordering events distributed algorithm given synchronizing logical clocks used totally order events use total ordering illustrated method solving synchronization problems algorithm specialized synchronizing physical clocks bound derived far synchrony clocks

#* distributed routing algorithm mobile radio networks
#@M. Scott Corson,Anthony Ephremides
#t1995
#cWireless Networks
#index139
#id139
#%
#! present distributed routing protocol mobile packet radio networks protocol intended use networks rate topological change fast make flooding possible routing method slow make existing protocols static topology applicable roue ing algorithm adapts asynchronously distributed fashion arbitrary changes topology absence global topological knowledge protocol maintains set loop free routes destination node desires route protocol performance measured terms end end packet delay throughput compared pure flooding alternative algorithm suited mediumrate topological change environment envisioned results rate topological changes high flooding preferable alternatives lower rates change appears effects channel access account ed performance new algorithm encouraging generally superior alternative protocols

#* holistic lexicon based approach opinion mining
#@Xiaowen Ding,Bing Liu,Philip S. Yu
#t2008
#cWSDM
#index140
#id140
#%141,142,143,144,145,146,147,148,149,150,151,152,153,154,155
#! important types information web opinions expressed user generated content customer reviews products forum posts blogs paper focus customer reviews products particular study problem determining semantic orientations positive negative neutral opinions expressed product features reviews problem applications opinion mining summarization search existing techniques utilize list opinion bearing words called opinion lexicon purpose opinion words words express desirable great amazing undesirable bad poor states approaches major shortcomings paper propose holistic lexicon based approach solving problem exploiting external evidences linguistic conventions natural language expressions approach allows handle opinion words context dependent cause major difficulties existing algorithms deals special words phrases language constructs impacts opinions based linguistic patterns effective function aggregating multiple conflicting opinion words sentence called opinion observer based proposed technique implemented experimental results using benchmark product review data set additional reviews proposed technique highly effective outperforms existing methods significantly

#* mining comparative sentences relations
#@Nitin Jindal,Bing Liu
#t2006
#cAAAI
#index141
#id141
#%
#! paper studies text mining problem comparative sentence mining csm comparative sentence expresses ordering relation sets entities respect common features example comparative sentence canon optics better sony nikon expresses comparative relation better given set evaluative texts web reviews forum postings news articles task comparative sentence mining identify comparative sentences texts extract comparative relations identified comparative sentences problem applications example product manufacturer wants know customer opinions products comparison competitors paper propose novel techniques based new types sequential rules perform tasks experimental evaluation conducted using different types evaluative texts web results techniques promising

#* automatic construction polarity tagged corpus html documents
#@Nobuhiro Kaji,Masaru Kitsuregawa
#t2006
#cACL
#index142
#id142
#%
#! paper proposes novel method building polarity tagged corpus html documents characteristics method fully automatic applied arbitrary html documents idea method utilize certain layout structures linguistic pattern using automatically extract sentences express opinion experiment method construct corpus consisting sentences

#* determining sentiment opinions
#@Soo-Min Kim
#t2004
#cProceedings of COLING
#index143
#id143
#%
#! identifying sentiments affective parts opinions challenging problem present given topic automatically finds people hold opinions topic sentiment opinion contains module determining word sentiment combining sentiments sentence experiment various models classifying combining sentiment word sentence levels promising results

#* automatic identification pro reasons online reviews
#@Soo-Min Kim,Eduard H. Hovy
#t2006
#cACL
#index144
#id144
#%
#! paper present automatically extracts pros cons online reviews approaches developed extracting opinions text focus extracting reasons opinions form fact opinion leveraging online review sites author generated pros cons propose aligning pros cons sentences review texts maximum entropy model trained resulting labeled set subsequently extract pros cons online review sites explicitly provide experimental results resulting identifies pros cons precision recall

#* opinion extraction summarization tracking news blog corpora
#@Lun-Wei Ku,Yu-Ting Liang,Hsin-Hsi Chen
#t2006
#cProceedings of AAAI-2006 Spring Symposium on Computational Approaches to Analyzing Weblogs
#index145
#id145
#%
#! humans like express opinions eager know opinions automatically mining organizing opinions heterogeneous information sources useful individuals organizations governments opinion extraction opinion summarization opinion tracking important techniques understanding opinions opinion extraction mines opinions word sentence document levels articles opinion summarization summarizes opinions articles telling sentiment polarities degree correlated events paper news web blog articles investigated trec ntcir articles collected web blogs serve information sources opinion extraction documents related issue animal cloning selected experimental materials algorithms opinion extraction word sentence document level proposed issue relevant sentence selection discussed topical opinionated information summarized opinion summarizations visualized representative sentences text based summaries different languages different sources compared finally opinionated curve showing supportive nonsupportive degree timeline illustrated opinion tracking

#* opinion observer analyzing comparing opinions web
#@Bing Liu,Minqing Hu,Junsheng Cheng
#t2005
#cWWW
#index146
#id146
#%
#! web excellent source gathering consumer opinions numerous web sites containing opinions customer reviews products forums discussion groups blogs paper focuses online customer reviews products makes contributions proposes novel framework analyzing comparing consumer opinions competing products prototype called opinion observer implemented single glance visualization user able clearly strengths weaknesses product minds consumers terms various product features comparison useful potential customers product manufacturers potential customer visual feature feature comparison consumer opinions products helps decide product buy product manufacturer comparison enables easily gather marketing intelligence product benchmarking information second new technique based language pattern mining proposed extract product features pros cons particular type reviews features form basis comparison experimental results technique highly effective outperform existing methods significantly

#* examining role linguistic knowledge sources automatic identification classification reviews
#@Vincent Ng,Sajib Dasgupta,S. M. Niaz Arifin
#t2006
#cACL
#index147
#id147
#%
#! paper examines problems document level sentiment analysis determining given document review classifying polarity review positive negative demonstrate review identification performed high accuracy using unigrams features examine role types simple linguistic knowledge sources polarity classification

#* seeing stars exploiting class relationships sentiment categorization respect rating scales
#@Bo Pang,Lillian Lee
#t2005
#cACL
#index148
#id148
#%
#! address rating inference problem simply decide review thumbs thumbs previous sentiment analysis work determine author evaluation respect multi point scale stars task represents interesting twist standard multi class text categorization different degrees similarity class labels example stars intuitively closer stars star evaluate human performance task apply meta algorithm based metric labeling formulation problem alters given ary classifier output explicit attempt ensure similar items receive similar labels meta algorithm provide significant improvements multi class regression versions svms employ novel similarity measure appropriate problem

#* thumbs sentiment classification using machine learning techniques
#@Bo Pang,Lillian Lee,Shivakumar Vaithyanathan
#t2002
#cCoRR
#index149
#id149
#%
#! consider problem classifying documents topic overall sentiment determining review positive negative using movie reviews data standard machine learning techniques definitively outperform human produced baselines machine learning methods employed naive bayes maximum entropy classification support vector machines perform sentiment classification traditional topic based categorization conclude examining factors make sentiment classification problem challenging

#* extracting product features opinions reviews
#@Ana-Maria Popescu,Oren Etzioni
#t2005
#cHLT/EMNLP
#index150
#id150
#%
#! consumers forced wade line reviews order make informed product choice paper introduces opine unsupervised information extraction mines reviews order build model important product features evaluation reviewers relative quality products compared previous work opine achieves higher precision lower recall feature extraction task opine novel use relaxation labeling finding semantic orientation words context leads strong performance tasks finding opinion phrases polarity

#* learning extraction patterns subjective expressions
#@Ellen Riloff,Ellen Riloff
#t2003
#cProceedings of the 2003 Conference on Empirical Methods in Natural Language Processing
#index151
#id151
#%
#! paper presents bootstrapping process learns linguistically rich extraction patterns subjective opinionated expressions high precision classifiers label unannotated data automatically create large training set given extraction pattern learning algorithm learned patterns used identify subjective sentences bootstrapping process learns subjective patterns increases recall maintaining high precision

#* opinion summarization linking sources
#@Veselin Stoyanov,Claire Cardie
#t2006
#cWorkshop On Sentiment And Subjectivity In Text
#index152
#id152
#%
#! target problem linking source mentions belong entity source coreference resolution needed creating opinion summaries paper source coreference resolution transformed standard noun phrase coreference resolution apply state art coreference resolution approach transformed data evaluate available corpus manually annotated opinions

#* word sense subjectivity
#@Janyce Wiebe,Rada Mihalcea
#t2006
#cACL
#index153
#id153
#%
#! subjectivity meaning important properties language paper explores interaction brings empirical evidence support hypotheses subjectivity property associated word senses word sense disambiguation directly benefit subjectivity annotations

#* answering opinion questions separating facts opinions identifying polarity opinion sentences
#@Hong Yu,Vasileios Hatzivassiloglou
#t2003
#cConference on Empirical Methods in Natural Language Processing
#index154
#id154
#%
#! opinion question answering challenging task natural language processing paper discuss necessary component opinion question answering separating opinions fact document sentence level present bayesian classifier discriminating documents preponderance opinions editorials regular news stories unsupervised statistical techniques significantly harder task detecting opinions sentence level present model classifying opinion sentences positive negative terms main perspective expressed opinion results large collection news stories human evaluation sentences reported indicating achieve high performance document classification upwards precision recall respectable performance detecting opinions classifying sentence level positive negative neutral accuracy

#* movie review mining summarization
#@Li Zhuang,Feng Jing,Xiaoyan Zhu
#t2006
#cCIKM
#index155
#id155
#%
#! flourish web online review useful important information resource people result automatic review mining summarization hot research topic recently different traditional text summarization review mining summarization aims extracting features reviewers express opinions determining opinions positive negative paper focus specific domain movie review multi knowledge based approach proposed integrates wordnet statistical analysis movie knowledge experimental results effectiveness proposed approach movie review mining summarization

#* microeconomic data mining problem customer oriented catalog segmentation
#@Martin Ester,Rong Ge,Wen Jin,Zengjian Hu
#t2004
#cKDD
#index156
#id156
#%157,158,159,160,161,162,163,164,165,166
#! microeconomic framework data mining assumes enterprise chooses decision maximizing overall utility customers contribution customer function data available customer catalog segmentation enterprise wants design product catalogs size maximize overall number catalog products purchased applications customer attracted enterprise purchase products ones contained catalog paper investigate alternative problem formulation customer oriented catalog segmentation overall utility measured number customers specified minimum catalogs formally introduce customer oriented catalog segmentation problem discuss complexity investigate different paradigms design efficient approximate algorithms customer oriented catalog segmentation problem greedy deterministic randomized algorithms greedy algorithms trapped local optimum randomized algorithms crucially depend reasonable initial solution explore combination paradigms experimental evaluation synthetic real data demonstrates new algorithms yield catalogs significantly higher utility compared classical catalog segmentation algorithms

#* data mining framework optimal product selection retail supermarket data generalized profset model
#@Tom Brijs,Bart Goethals,Gilbert Swinnen,Koen Vanhoof,Geert Wets
#t2001
#cCoRR
#index157
#id157
#%
#! recent years data mining researchers developed efficient association rule algorithms retail market basket analysisfi retailers complain adopt association rules optimize concrete retail marketing mix decisionsfi context previous paper authors introduced product selection model called profsetfi model selects interesting products product assortment based cross selling potential given

#* threshold ln approximating set cover
#@Uriel Feige
#t1998
#cJ. ACM
#index158
#id158
#%
#! given collectionof subsets set cover problem selecting asfew possible subsets fromsuch union covers max cover problem ofselecting subsets fromsuch union maximum cardinality problems arenp hard prove ln threshold set cover beapproximated efficiently unless np slightly superpolynomial time algorithms closes gap low order terms ratio approximation achievable greedy algorithm ln previous results lund yannakakis showed hardness ofapproximation ratio log ln max cover approximationthreshold low order terms assumption pnp categories subject descriptors nonnumerical algorithms problems ccomputations discrete structures

#* microeconomic view data mining
#@Jon M. Kleinberg,Christos H. Papadimitriou,Prabhakar Raghavan
#t1998
#cData Min. Knowl. Discov.
#index159
#id159
#%
#! present rigorous framework based optimization evaluating data miningoperations associations clustering terms utility decisionmakingfithis framework leads quickly interesting computational problemsrelated sensitivity analysis segmentation theory gamesfidepartment science cornell university ithaca ny fi email kleinber csficornellfiedufisupported alfred pfi sloan research fellowship nsf

#* mining values added association rules
#@Rakesh Agrawal,Tomasz Imielinski,Arun Swami
#t1993
#cACM SIGMOD Record
#index160
#id160
#%
#! value added product industrial term referring minor addition major products paper borrow term denote minor semantic addition known association rules consider addition numerical values attribute values sale price profit degree fuzziness level security additions lead notion random variables added value attributes data model probabilistic considerations data mining

#* theoretical framework data mining
#@Heikki Mannila
#t2000
#cACM SIGKDD Explorations Newsletter
#index161
#id161
#%
#! research data mining knowledge discovery databases concentrated developing good algorithms various data mining tasks example recent proceedings kdd conferences parts research effort gone investigating data mining process user interface issues database topics visualization relatively little published theoretical foundations data mining paper present possible theoretical approaches data mining area infancy probably questions answers paper

#* efficient algorithms creating product catalogs
#@Michael Steinbach,G Karypis,Vipin Kumar
#t2001
#cProc. of. SDM 2001
#index162
#id162
#%
#! purposes paper define catalog promotional catalog collection products items presented customer hope encouraging purchase single mailing problem addresses build collection catalogs distribute customers customer achieve optimal outcome profit catalog subset given set items different catalogs contain items catalogs overlap slightly general important extension single mailing problem seeks optimal set catalogs multiple mailings allowed multiple catalogs sent customer catalog creation important applications commerce traditional brick mortar retailers especially used personalized recommender systems catalog creation problem np complete relatively expensive approximation algorithms recently developed paper efficient techniques building catalogs algorithms outperform previously suggested approaches techniques previously suggested feasible realistic numbers customers catalogs techniques directly use objective function maximize profit locally optimal solution approach based gradient ascent combining techniques clustering similar customers better results obtained analyze performance algorithms respect theoretical bound cases performance close optimal

#* mpis maximal profit item selection cross selling considerations
#@Raymond Chi-Wing Wong,Ada Wai-Chee Fu,Ke Wang
#t2003
#cICDM
#index163
#id163
#%11,12,167,168,157,169,170,171,172,173,174,159,175,176,177,178,179,180
#! literature data mining different algorithms association rule mining proposed relatively little study association rules aid specific targets paper applications association rules maximal profit item selection cross selling effect mpis problem investigated problem selecting subset items maximal profit consideration cross selling prove simple version problem np hard propose new approach problem consideration loss rule kind association rule model cross selling effect problem transformed quadratic programming problem case quadratic programming applicable propose heuristic approach experiments conducted proposed methods highly effective efficient

#* item selection hub authority profit ranking
#@Ke Wang,Ming-Yen Thomas Su
#t2002
#cInternational Conference on Knowledge Discovery and Data Mining
#index164
#id164
#%
#! fundamental problem business applications ranking items respect notion profit based historical transactions difficulty profit item comes sales influence sales items cross selling effect paper draw analogy influence mutual reinforcement hub authority web pages based analogy present novel approach item

#* approximate catalog segmentation problem using semidefinite programming relaxations
#@Dachuan Xu,Yinyu Ye b,Jiawei Zhang c
#t2002
#cOptimization Method and Software
#index165
#id165
#%
#! consider catalog segmentation problem csp introduced kleinberg papadimitriou raghavan cite kpr fi problem given ground set items family cdots subsets integer le le fi objective subsets subset sum max cap cap maximizedfi known simple greedy algorithm performance guarantee

#* profit mining patterns actions
#@Ke Wang,Senqiang Zhou,Jiawei Han
#t2002
#cEDBT
#index166
#id166
#%
#! major obstacle data mining applications gap betweenthe statistic based pattern extraction value based decisionmakingfi present profit mining approach reduce gapfi profitmining given set past transactions pre selected targetitems like build model recommending target items andpromotion strategies new customers goal maximizing thenet profitfi identify issues profit mining propose

#* heuristic algorithms unconstrained binary quadratic programming problem
#@J.E. Beasley
#t1998
#cTechnical report
#index167
#id167
#%
#! paper consider unconstrained binary quadratic programming problemfi problem maximising quadratic objective suitable choice binary zero variablesfi present heuristic algorithms based tabu search simulated annealing problemfi computational results presented number publically available data sets involving variablesfi interesting feature results whilst problems tabu search dominates

#* transaction tells story
#@
#t1000
#cEMPTY
#index168
#id168
#%
#! financial institutions retailers payment processors running point sale pos automated teller machine atm terminals looking ways lower operational costs increasing risk service disruption end customers utilizing dial terminals instead costly service terminals looking implement payment solution

#* using association rules product assortment decisions case study
#@Tom Brijs,Gilbert Swinnen,Koen Vanhoof,Geert Wets
#t1999
#cKDD
#index169
#id169
#%
#! claimed discovery association rules iswell suited applications market basket analysis revealregularities purchase behaviour customersfi recent work indicates discovery interesting rules canin fact addressed microeconomic frameworkfithis study integrates discovery frequent itemsets microeconomic model product selection profset model enable integration quantitative

#* computers intractability guide theory np completeness
#@M. R. Garey,David S. Johnson
#t1979
#cnull
#index170
#id170
#%
#! book theory np completeness computational intractability book features appendix providing thorough compendium np complete problems updated later editions book book outdated respects does cover recent development pcp theorem print regarded classic study citeseer search engine listed book cited reference science literature

#* mining frequent patterns candidate generation
#@Jiawei Han,Jian Pei,Yiwen Yin
#t2000
#cSIGMOD Conference
#index171
#id171
#%
#! mining frequent patterns transaction databases time series databases kinds ofdatabases studied popularly data mining research previous studies adopt apriori likecandidate set generation test approach candidate set generation costly especially exist large number patterns long patterns study propose novelfrequent pattern tree fp tree structure extended prefix treestructure storing compressed crucial information frequent patterns develop efficient fp tree based mining method fp growth mining complete set frequent patterns pattern fragment growth efficiency mining achieved techniques large database compressed condensed smaller data structure fp tree avoids costly repeated database scans fp tree based mining adoptsa pattern fragment growth method avoid costly generation large number candidate sets apartitioning based divide conquer method used decompose mining task set smaller tasks formining confined patterns conditional databases dramatically reduces search space performancestudy shows fp growth method efficient scalable mining long short frequent patterns order magnitude faster apriori algorithm faster recently reportednew frequent pattern mining methods

#* finite algorithm maximize certain pseudoconcave functions polytopes
#@B. von Hohenbalken
#t1975
#cMathematical Programming
#index172
#id172
#%
#! paper develops proves algorithm finds exact maximum certain nonlinear functions polytopes performing finite number logical arithmetic operations permissible objective functions need pseudoconcave allow closed form solution sets equations order conditions associated unconstrained affinely transformed objective function examples pseudoconcave quadratics especially homogeneous functioncx xvx positive definite sofar finite algorithm existed distinction available methods algorithm uses internal representation feasible set selectively decompose simplices varying dimensions linear programming gradient criterion used select sequence simplices contain corresponding sequence strictly increasing relative relatively interior maxima greatest shown global maximum feasible set interior maxima simplices finite way calculus maximizations affine hulls subsets vertices necessary requirement explicitly solvable paper presents flow structure algorithm supporting theory decision theoretic use example computed apl version method

#* introduction global optimization
#@R Horst,PM Pardalos,NV Thoai
#t1995
#cNonconvex optimization and its applications
#index173
#id173
#%
#! accurate modelling real world problems requires nonconvex terms introduced model objective function constraints nonconvex programming thehardest fields optimization presenting challenges practical theoretical aspects presence multiple local minima calls application global optimization techniques paper mini course global optimization techniques nonconvex programming dealswith theoretical aspects nonlinear programming current state art algorithms global optimization syllabus follows examples nonlinearprogramming problems nlps general description phase algorithms local optimizationof nlps derivation kkt conditions short notes stochastic global multistart algorithmswith concrete example sobolopt depth study deterministic spatial branch boundalgorithm convex relaxation nlp latest advances bilinear programming theory ofreduction constraints

#* quadratic binary programming dynamical approach determine predictability epileptic seizures
#@Leonidas D. Iasemidis,PM Pardalos,J. Chris Sackellares,Deng-Shan Shiau
#t2001
#cJ. Comb. Optim.
#index174
#id174
#%
#! epilepsy common disorders nervous progressive entrainment epileptogenic focus normal brain areas results transitions brain chaotic chaotic spatiotemporal states epileptic seizures entrainment brain sites quantified index measures chaos lyapunov exponents electrical activity eeg brain applying optimization theory particular quadratic zero programming able select entrained brain sites minutes seizures subsequently follow entrainment hours seizures patients seizures seizures predictable optimal selection electrode sites procedure applied epilepsy research time shows possibility prediction epileptic seizures advance minutes occurrence

#* authoritative sources hyperlinked environment
#@Jon M. Kleinberg
#t1999
#cJ. ACM
#index175
#id175
#%
#! network structure hyperlinked environment rich source formation content environment provided effective means forunderstanding develop set algorithmic tools extracting information fromthe link structures environments report experiments demonstratetheir effectiveness variety contexts world wide web central issuewe address framework distillation broad search topics throughthe discovery authoritative information sources topics propose andtest algorithmic formulation notion authority based relationshipbetween set relevant authoritative pages set hub pages join link structure formulation connections eigenvectorsof certain matrices associated link graph connections turn motivateadditional heuristics link based analysis

#* methods problems data mining
#@Heikki Mannila
#t1997
#cICDT
#index176
#id176
#%
#! knowledge discovery databases data mining aim semiautomatic tools analysis large data setsfi consider methods used data mining concentrating levelwise search frequently occurring patternsfi technique used various applicationsfi discuss possibilities compiling data mining queries algorithms look use sampling data miningfi conclude listing open research problems data mining

#* efficient algorithms discovering association rules
#@Heikki Mannila,Hannu Toivonen,A. Inkeri Verkamo
#t1994
#cKDD Workshop
#index177
#id177
#%
#! association rules statements form rows relation row value columns set column fi agrawal imielinski swami introduced problem mining association rules large collections data gave method based successive passes database improved algorithm problemfi method based careful combinatorial analysis information obtained previous passes

#* optimizing web servers using page rank prefetching clustered accesses
#@Victor Safronov,Manish Parashar
#t2003
#cInf. Sci.
#index178
#id178
#%
#! paper presents page rank based prefetching technique accesses web page clusters approach uses link structure requested page determine important linked pages identify page prefetched underlying premise approach case cluster accesses pages requested users web server typically based current previous pages requested furthermore requested pages lot links important page page higher probability requested experimental evaluation prefetching mechanism presented using real server logs results page rank based scheme does better random prefetching clustered accesses hit rates cases

#* computationally related problems
#@Sartaj Sahni
#t1974
#cSIAM J. Comput.
#index179
#id179
#%
#! look problems areas network flows game theory artificial intelligence graph theory integer programming nonlinear programming related problems solvable polynomial time present polynomial time algorithm problems known problems extend equivalence class problems known complete problem deciding class languages accepted polynomial time nondeterministic turing machines accepted polynomial time deterministic turing machines related complete problems classes languages complete problem polynomial deterministic solution view appears likely equivalence class defines class problems solved deterministic polynomial time

#* item selection hub authority profit ranking
#@
#t0
#c
#index180
#id180
#%
#! fundamental problem business applications ranking items respect notion profit based historical transactions difficulty profit item comes sales influence sales items cross selling effect paper draw analogy influence mutual reinforcement hub authority web pages based analogy present novel approach item

#* sentiment aware model predicting sales
#@Yang Liu,Xiangji Huang,Aijun An,Xiaohui Yu
#t2007
#cSIGIR '07
#index181
#id181
#%182,183,184,185,186,187,188,189,190,191,192,146,193,194,195,148,149
#! high popularity weblogs blogs short present wealth information helpful assessing general public sentiments opinions paper study problem mining sentiment information blogs investigate ways use information predicting product sales performance based analysis complex nature sentiments propose sentiment plsa plsa blog entry viewed document generated number hidden sentiment factors training plsa model blog data enables obtain succinct summary sentiment information embedded blogs present arsa autoregressive sentiment aware model utilize sentiment information captured plsa predicting product sales performance extensive experiments conducted movie data set compare arsa alternative models account sentiment information model different feature selection method experiments confirm effectiveness superiority proposed approach

#* latent dirichlet allocation
#@David M. Blei,Andrew Y. Ng,Michael I. Jordan
#t2003
#cJournal of Machine Learning Research
#index182
#id182
#%196,197,198,199,66,200,201,202,203,204,205,206,207,208,209,210,211,212,213,214,215,216,85
#! latent dirichlet allocation lda generative probabilistic model collections discrete data text corpora lda level hierarchical bayesian model item collection modeled finite mixture underlying set topics topic turn modeled infinite mixture underlying set topic probabilities context text modeling topic probabilities provide explicit representation document present efficient approximate inference techniques based variational methods em algorithm empirical bayes parameter estimation report results document modeling text classification collaborative filtering comparing mixture unigrams model probabilistic lsi model

#* spatio temporal analysis online news blogs
#@Angelo Dalli
#t2006
#cWWW
#index183
#id183
#%
#! previous work spatio temporal analysis news items documents largely focused broad categorization small text collections region country large scale spatio temporal analysis online news media blogs presented analysis global news media coverage year period demonstrate benefits using hierarchical geospatial database disambiguate geographical named entities provide results extremely fine grained analysis news items aggregate maps media attention particular places world compared geographical socio economic data analysis suggests gdp capita best indicator media attention

#* liberal media right wing conspiracies using cocitation information estimate political orientation web documents
#@Miles Efron
#t2004
#cCIKM
#index184
#id184
#%
#! paper introduces simple method estimating cultural orientation affiliation online entities polarized field discourse particular cocitation information used estimate political orientation hypertext documents type cultural orientation political orientation document degree participates traditionally left right wing beliefs estimating documents political orientation personalized information retrieval recommender systems application politics method uses simple probabilistic model estimate strength association document left right wing communities model estimates likelihood cocitation document small number documents known orientation model tested sets data partisan web documents political weblogs non partisan documents accuracy obtained cocitation model outperforming lexically based classifiers statistically significant levels

#* predictive power online chatter
#@Daniel Gruhl,Ramanathan V. Guha,Ravi Kumar,Jasmine Novak,Andrew Tomkins
#t2005
#cKDD
#index185
#id185
#%
#! increasing fraction global discourse migrating online form blogs bulletin boards web pages wikis editorials dizzying array new collaborative technologies migration proceeded point topics reflecting certain individual products sufficiently popular allow targeted online tracking ebb flow chatter topics based analysis half million sales rank values books period months correlating postings blogs media web pages able draw interesting conclusions carefully hand crafted queries produce matching postings volume predicts sales ranks second queries automatically generated cases sales rank motion difficult predict general algorithmic predictors use online postings successfully predict spikes sales rank

#* information diffusion blogspace
#@Daniel Gruhl,Ramanathan V. Guha,David Liben-Nowell,Andrew Tomkins
#t2004
#cWWW
#index186
#id186
#%
#! study dynamics information propagation environments low overhead personal publishing using large collection weblogs time example domain characterize model collection levels present macroscopic characterization topic propagation corpus formalizing notion long running chatter topics consisting recursively spike topics generated outside world events rarely resonances community second present microscopic characterization propagation individual individual drawing theory infectious diseases model flow propose validate employ algorithm induce underlying propagation network sequence posts report results

#* probabilistic latent semantic analysis
#@Thomas Hofmann
#t1999
#cUAI
#index187
#id187
#%
#! probabilistic latent semantic analysis novel statistical technique analysis mode occurrence data applications information retrieval ltering natural language processing ma chine learning text related ar easfi compared standard latent semantic analysis stems linear algebra performs singular value decomposition occurrence tables proposed method based mixture decomposition derived latent class modelfi results principled approach solid foundation statisticsfi order avoid tting propose widely applicable generalization maximum likelihood model tting tempered emfi approach yields substantial consistent improvements latent semantic analysis number ex perimentsfi

#* dynamic real time forecasting online auctions functional models
#@Wolfgang Jank,Galit Shmueli,Shanshan Wang
#t2006
#cKDD
#index188
#id188
#%
#! propose dynamic model forecasting price online auctions key features model operates live auction makes different previous approaches consider static models model different respect information price incorporated model based traditional notion auction price level incorporates dynamics form price velocity acceleration sense incorporates key features dynamic environment online auction use novel functional data methodology allows measure subsequently include dynamic price characteristics illustrate model diverse set ebay auctions different book categories significantly higher prediction accuracy compared standard approaches

#* words attitude
#@Jaap Kamps,Maarten Marx
#t2001
#c1st International WordNet Conference
#index189
#id189
#%
#! traditional notion word meaning used natural language processing literal lexical meaning used dictionaries lexicons relatively objective notion lexical meaning different subjective notions emotive affective meaning aim come grips subjective aspects meaning expressed written texts attitude value expressed paper explores structure wordnet lexical database used assess affective emotive meaning particular construct measures based osgood semantic differential technique

#* bursty evolution blogspace
#@Ravi Kumar,Jasmine Novak,Prabhakar Raghavan,Andrew Tomkins
#t2005
#cWorld Wide Web
#index190
#id190
#%
#! propose new tools address evolution hyperlinked corpora define time graphs extend traditional notion evolving directed graph capturing link creation point phenomenon time second develop definitions algorithms time dense community tracking crystallize notion community evolution develop tools context blogspace space weblogs blogs study involves approximately links blogs create time graph blogs automatic analysis internal time stamps study evolution connected component structure microscopic community structure time graph blogspace underwent transition behavior end rapidly expanding past year just metrics scale metrics community structure connectedness expansion shows sign abating measures connectedness plateau years randomizing link destinations blogspace retaining sources timestamps introduce concept randomized blogspace observe similar evolution giant component corresponding increase community structure having demonstrated formation micro communities time turn ongoing activity active communities extend recent work kleinberg discover dense periods bursty intra community link creation

#* structure evolution blogspace
#@Ravi Kumar,Jasmine Novak,Prabhakar Raghavan,Andrew Tomkins
#t2004
#cCommun. ACM
#index191
#id191
#%
#! critical look million bloggers individual entries blogs reveals blogger demographics friendships activity patterns time

#* probabilistic model retrospective news event detection
#@Zhiwei Li,Bin Wang,Mingjing Li,Wei-Ying Ma
#t2005
#cSIGIR
#index192
#id192
#%
#! retrospective news event detection red defined discovery previously unidentified events historical news corpus contents time information news articles helpful red researches focus utilization contents news articles research works carried finding better usages time information paper explorations directions based following characteristics news articles hand news articles aroused events hand similar articles reporting event redundantly appear news sources hints generative model news articles provides data enriched environments perform red consideration characteristics propose probabilistic model incorporate content time information unified framework model gives new representations news articles news events furthermore based approach build interactive red hiscovery provides additional functions present events photo story chronicle

#* probabilistic approach spatiotemporal theme pattern mining weblogs
#@Qiaozhu Mei,Chao Liu,Hang Su,ChengXiang Zhai
#t2006
#cWWW
#index193
#id193
#%
#! mining subtopics weblogs analyzing spatiotemporal patterns applications multiple domains paper define novel problem mining spatiotemporal theme patterns weblogs propose novel probabilistic approach model subtopic themes spatiotemporal theme patterns simultaneously proposed model discovers spatiotemporal theme patterns extracting common themes weblogs generating theme life cycles given location generating theme snapshots given time period evolution patterns discovered comparative analysis theme life cycles theme snapshots experiments different data sets proposed approach discover interesting spatiotemporal theme patterns effectively proposed probabilistic model general used spatiotemporal text mining domain time location information

#* mixture model contextual text mining
#@Qiaozhu Mei,ChengXiang Zhai
#t2006
#cKDD
#index194
#id194
#%
#! contextual text mining concerned extracting topical themes text collection context information time location comparing analyzing variations themes different contexts topics covered document usually related context document analyzing topical themes context potentially reveal interesting theme patterns paper generalize models proposed previous work propose new general probabilistic model contextual text mining cover existing models special cases specifically extend probabilistic latent semantic analysis plsa model introducing context variables model context document proposed mixture model called contextual probabilistic latent semantic analysis cplsa model applied interesting mining tasks temporal text mining spatiotemporal text mining author topic analysis cross collection comparative analysis empirical experiments proposed mixture model discover themes contextual variations effectively

#* sentimental education sentiment analysis using subjectivity summarization based minimum cuts
#@Bo Pang,Lillian Lee
#t2004
#cACL
#index195
#id195
#%
#! sentiment analysis seeks identify viewpoint underlying text span example application classifying movie review thumbs thumbs determine sentiment polarity propose novel machine learning method applies text categorization techniques just subjective portions document extracting portions implemented using efficient techniques finding minimum cuts graphs greatly facilitates incorporation cross sentence contextual constraints

#* variational bayesian framework graphical models
#@H Attias
#t2000
#cAdvances in Neural Information Processing Systems 12
#index196
#id196
#%
#! paper presents novel practical framework bayesian modelaveraging model selection probabilistic graphical modelsfiour approach approximates posterior distributions modelparameters structures latent variables analyticalmannerfi posteriors fall free form optimizationprocedure naturally incorporates conjugate priorsfi unlikein large sample approximations posteriors generally nongaussianand hessian needs

#* modern information retrieval
#@Ricardo A. Baeza-Yates,Berthier A. Ribeiro-Neto
#t1999
#cnull
#index197
#id197
#%
#! information retrieval ir changed considerably years expansion web world wide web advent modern inexpensive graphical user interfaces mass storage devices result traditional ir textbooks quite date led introduction new ir books recently believe great need book approaches field rigorous complete way science perspective opposition user centered perspective book effort partially fulfill gap useful course information retrieval graduate course topic www pages digital version book complete contents preface table contents glossary chapters available reading line printed version ordered directly addison wesley longman

#* modeling annotated data
#@David M. Blei,Michael I. Jordan
#t2003
#cSIGIR
#index198
#id198
#%196,375,182,676,677,678,679,680,681,271,266
#! consider problem modeling annotated datadata multiple types instance type caption serves description type image hierarchical probabilistic mixture models aim data culminating correspondence latent dirichlet allocation latent variable model ective modeling joint distribution types conditional distribution annotation given primary type conduct experiments corel database images captions assessing performance terms held likelihood automatic annotation text based image retrieval

#* theory probability
#@Harold Jeffreys
#t1998
#cbook
#index199
#id199
#%
#! probability theory branch mathematics concerned analysis random phenomena central objects probability theory random variables stochastic processes events mathematical ions non deterministic events measured quantities single occurrences evolve time apparently random fashion individual coin toss roll die random event repeated times sequence random events exhibit certain statistical patterns studied predicted representative mathematical results describing patterns law large numbers central limit theorem mathematical foundation statistics probability theory essential human activities involve quantitative analysis large sets data methods probability theory apply description complex systems given partial knowledge state statistical mechanics great discovery twentieth century physics probabilistic nature physical phenomena atomic scales described quantum mechanics

#* bayesian methods censored categorical data
#@James M. Dickey,Jhy-Ming Jiang, Joseph B. Kadane
#t1987
#cJournal of the American Statistical Association
#index200
#id200
#%
#! bayesian methods given finite category sampling observations suffer missing category distinctions dickey generalization dirichlet family prior distributions closed censored sampling posterior moments predictive probabilities proportional ratios carlson multiple hypergeometric functions closed form expressions developed case nested reported sets bayesian estimates computed easily relative frequencies effective computational methods given general case example involving surveys death penalty attitudes used illustrate theory simple special case categorical missing data way contingency table cross classified count data xsubij sub supplementary trials counted margin distinguishing rows ysubi sub supplementary trials reported counts distinguishing columns zsubj sub assumptions censoring process itself noninformative regarding category probabilities subij sub report possible outcome nonrandom prespecified bayesian inference regarding subij sub based likelihood function subij sub thetasupxsubij sub supsubij sub likelihood ordinarily considered intractable unsuited bayesian conjugate prior inference develop bayesian conjugate theory recognizing complete integrals functions carlson functions posterior distributions resulting dirichlet prior distributions known generalized dirichlet distributions corresponding posterior density functions similar form likelihood constitute family distributions closed sampling tractable various senses including convenient computability moments modes

#* bayesian data analysis
#@Andrew Gelman,John B. Carlin,Hal S. Stern,D. B. Rubin
#t2003
#cbook
#index201
#id201
#%
#! incorporating new updated information second edition bestselling text bayesian data analysis continues emphasize practice theory describing conceptualize perform critiques statistical analysis bayesian perspective changes new edition include added material bayesian methods connected approaches stronger focus mcmc added chapter computation topics examples additional chapters current models bayesian data analysis equation models generalized linear mixed models book introductory text reference working scientists professional life

#* probabilistic approach semantic representation
#@Thomas L. Griffiths,Mark Steyvers
#t2002
#cProceedings of the. 24th Annual Conference of the Cognitive Science Society
#index202
#id202
#%
#! semantic networks produced human data havestatistical properties easily capturedby spatial representationsfi explore probabilisticapproach semantic representation explicitlymodels probability words occurin di erent contexts captures probabilisticrelationships wordsfi thatthis representation statistical properties consistentwith large scale structure semantic networksconstructed humans

#* overview text retrieval conference trec
#@Donna Harman
#t1992
#cTREC
#index203
#id203
#%
#! text retrieval conference trec held early november attended people working participating groups goal conference bring research groups discuss work new large test collection large variety retrieval techniques reported including methods using automatic thesaurii sophisticated term weighting natural language techniques relevance feedback advanced pattern matching results run common evaluation package groups able compare effectiveness different techniques discuss differences sytems affected performance

#* experimental comparison clustering initialization methods
#@Marina Meila,D. Heckerman
#t1998
#cUAI
#index204
#id204
#%
#! examine methods clustering high dimensions paper perform experimental comparison batch clustering algorithms expectation maximization em algorithm winner version em algorithm reminiscent means algorithm model based hierarchical agglomerative clustering learn naive bayes models hidden root node using highdimensional discrete variable data sets real synthetic

#* probabilistic latent semantic indexing
#@Thomas Hofmann
#t1999
#cSIGIR
#index205
#id205
#%66,421,422,423,187,424,425,426,427,428,85,429,430
#! probabilistic latent semantic indexing novel approach automated document indexing based statistical latent class model factor analysis count data fitted training corpus text documents generalization expectation maximization algorithm utilized model able deal domain specific synonymy polysemous words contrast standard latent semantic indexing lsi singular value decomposition probabilistic variant solid statistical foundation defines proper generative data model retrieval experiments number test collections indicate substantial performance gains direct term matching methods lsi particular combination models different dimensionalities proven advantageous

#* statistical methods speech recognition
#@Frederick Jelinek
#t1998
#cbook
#index206
#id206
#%
#! book reflects decades important research mathematical foundations speech recognition focuses underlying statistical techniques hidden markov models decision trees expectation maximization algorithm information theoretic goodness criteria maximum entropy probability estimation parameter data clustering smoothing probability distributions author goal present principles clearly simplest setting advantages self organization real data enable reader apply techniques

#* making large scale svm learning practical
#@Thorsten Joachims
#t1998
#cAdvances in kernel methods: support vector learning book contents
#index207
#id207
#%
#! training support vector machine svm leads quadratic optimization problem bound constraints linear equality constraint despite fact type problem understood issues considered designing svm learner particular large learning tasks training examples shelf optimization techniques general quadratic programs quickly intractable memory time requirements svmlight

#* learning graphical models
#@Michael I. Jordan
#t1998
#cbook
#index208
#id208
#%
#! graphical models marriage probability theory graph theory provide natural tool dealing problems occur applied mathematics engineering uncertainty complexity particular play increasingly important role design analysis machine learning algorithms fundamental idea graphical model notion modularity complex built combining simpler parts probability theory serves glue parts combined ensuring consistent providing ways interface models data graph theory provides intuitively appealing interface humans model highly interacting sets variables data structure lends itself naturally design efficient general purpose algorithms book presents depth exploration issues related learning graphical model formalism chapters tutorial chapters robert cowell inference bayesian networks david mackay monte carlo methods michael jordan et al variational methods david heckerman learning bayesian networks remaining chapters cover wide range topics current research

#* introduction variational methods graphical models
#@Michael I. Jordan,Zoubin Ghahramani,Tommi Jaakkola,Lawrence K. Saul
#t1999
#cMachine Learning
#index209
#id209
#%
#! paper presents tutorial introduction use variational methods inference learningin graphical models bayesian networks markov random fields present number examples graphicalmodels including qmr dt database sigmoid belief network boltzmann machine variantsof hidden markov models infeasible run exact inference algorithms introduce variationalmethods exploit laws large numbers transform original graphical model simplified graphicalmodel inference efficient inference simpified model provides bounds probabilities interestin original model general framework generating variational transformations based convexduality finally return examples demonstrate variational algorithms formulated eachcase

#* general lower bounds based generated higher order expansions
#@Martijn A. R. Leisink,Hilbert J. Kappen
#t2002
#cUAI
#index210
#id210
#%
#! article rough outline algorithm generate lower bounds exponential function principle arbitrary precisionfi implemented generate necessary analytic terms boltzmann machine partition function leading lower bounds orderfi turns extra variational parameters optimized analyticallyfi bounds upto nineth order reasonably calculable practical situationsfi generated

#* expectation propagation generative aspect model
#@Thomas Minka,J Lafferty
#t2002
#cProceedings of the 18th Conference on Uncertainty in Artificial Intelligence
#index211
#id211
#%
#! generative aspect model extension multinomial model text allows word probabilities vary stochastically documents

#* using maximum entropy text classification
#@K Nigam,J Lafferty,A McCallum
#t1000
#cEMPTY
#index212
#id212
#%
#! paper proposes use maximum entropytechniques text classificationfi maximumentropy probability distribution estimationtechnique widely used variety ofnatural language tasks language modeling speech tagging text segmentationfithe underlying principle maximumentropy external knowledge oneshould prefer distributions uniformficonstraints distribution derived fromlabeled training data inform maximum entropy

#* text classification labeled unlabeled documents using em
#@K Nigam,A McCallum,Sebastian Thrun,Tom M. Mitchell
#t2000
#cMachine Learning
#index213
#id213
#%
#! paper shows accuracy learned text classifiers improved byaugmenting small number labeled training documents large pool unlabeled docu ments important text classification problems obtaining training labelsis expensive large quantities unlabeled documents readily available introduce algorithm learning labeled unlabeled documents based combination expectation maximization em naive bayes classifier algorithm firsttrains classifier using available labeled documents probabilistically labels unlabeleddocuments trains new classifier using labels documents iteratesto convergence basic em procedure works data conform generativeassumptions model assumptions violated practice poorperformance result present extensions algorithm improve classificationaccuracy conditions weighting factor modulate contribution theunlabeled data use multiple mixture components class experimental results obtained using text different real world tasks use unlabeled datareduces classification error

#* latent semantic indexing probabilistic analysis
#@Christos H. Papadimitriou,Prabhakar Raghavan,Hisao Tamaki,Santosh Vempala
#t2000
#cJ. Comput. Syst. Sci.
#index214
#id214
#%
#! latent semantic indexing lsi information retrieval technique based spectralanalysis term document matrix empirical success heretofore withoutrigorous prediction explanationfi prove certain conditions lsi does succeedin capturing underlying semantics corpus achieves improved retrieval performancefiwe propose technique random projection way speeding lsifi complementour theorems

#* probabilistic models unified collaborative content based recommendation sparse data environments
#@Alexandrin Popescul,Lyle H. Ungar,DM Pennock,S Lawrence
#t2001
#cUAI
#index215
#id215
#%
#! recommender systems leverage product community information target products consumersfi researchers developed collaborative recommenders content based recommenders hybrid systemsfi propose unified probabilistic framework merging collaborative content based recommendationsfi extend hofmann aspect model incorporate way occurrence data users items item contentfi relative influence collaboration data versus content data

#* improving multi class text classification naive bayes
#@Tommi Jaakkola,Jason D. M. Rennie,Jason D. M. Rennie
#t2001
#cAI Technical Reports
#index216
#id216
#%
#! numerous text documents available electronic formfi available dayfi documents represent massive information easily accessiblefi seeking value huge collection requires organization work organizing documents automated text classificationfi accuracy understanding systems greatly influences usefulnessfi paper seek advance understanding commonly

#* analysis weblog comments
#@G Mishne,N Glance 
#t2006
#cThird annual workshop on the Weblogging ecosystem
#index217
#id217
#%218,219,220,190,221,222,223,224,225,226,227
#! access weblogs commercial services academic studies usually limited content weblog posts overlooks important aspect distinguishing weblogs web pages ability weblog readers respond posts directly posting comments paper present large scale study weblog comments relation posts using sizable corpus comments estimate overall volume comments blogosphere analyze relation weblog popularity commenting patterns measure contribution comment content various aspects weblog access

#* argumentation analysis weblog conversations
#@A. de Moor,L. Efimova
#t0
#c
#index218
#id218
#%
#! weblogs important new components internet provide individual users easy way publish online comment views furthermore suite secondary applications allow weblogs linked searched navigated originally intended individual use practice weblogs increasingly appear facilitate distributed conversations important implications use technology medium collaboration given special characteristics weblogs supporting applications suited range conversational purposes require different forms argumentation paper analyze argumentation potential weblog technologies using diagnostic framework argumentation technologies pay special attention conversation structures dynamics weblogs naturally afford based initial analysis make number recommendations research apply technologies purposeful conversation processes knowledge management

#* blogpulse automated trend discovery weblogs
#@N Glance,Matthew Hurst,Takashi Tomokiyo
#t2004
#cWWW 2004 Workshop on the Weblogging Ecosystem
#index219
#id219
#%
#! past years weblogs emerged new communication publication medium internet paper application data mining information extraction nlp algorithms discovering trends subset approximately weblogs publish daily lists key persons key phrases key paragraphs public web site blogpulse com addition maintain searchable index weblog entries search index implemented trend search graphs normalized trend line time search query provides way estimate relative buzz word mouth given topics time

#* bridging gap genre analysis weblogs
#@Susan C. Herring,Lois Ann Scheidt,Sabrina Bonus,Elijah Wright
#t2004
#cHICSS
#index220
#id220
#%
#! weblogs blogs frequently modified web pages dated entries listed reverse chronological sequence latest genre internet communication attain widespread popularity characteristics systematically described paper presents results content analysis randomly selected weblogs comparing empirically observable features corpus popular claims nature weblogs finding differ number respects notably blog authors journalists scholars alike exaggerate extent blogs interlinked interactive oriented external events underestimate importance blogs individualistic intimate forms self expression based profile generated empirical analysis consider likely antecedents blog genre situate respect dominant forms digital communication internet today advance predictions long term impacts

#* audience structure authority weblog community
#@Cameron Marlow
#t2004
#c54th Annual Conference of the International Communications Association
#index221
#id221
#%
#! weblog medium fundamentally innovation personal publishing come engender new form social interaction web massively distributed completely connected conversation covering imaginable topic byproduct ongoing communication set hyperlinks weblogs exchange dialog form social acknowledgement authors paper seeks understand social implications linking community drawing hyperlink citations collected blogdex project past years social network analysis employed resulting social structure measures authority explored popularity measured webloggers public affiliations influence measured citation writing metrics evaluated respect authority conferred references popular press

#* blocking blog spam language model disagreement
#@G Mishne,David Carmel,Ronny Lempel
#t2005
#cAIRWeb
#index222
#id222
#%
#! present approach detecting link spam common blog comments comparing language models used blog post comment pages linked comments contrast link spam filtering approaches method requires training hard coded rule sets knowledge complete web connectivity preliminary experiments identification typical blog spam promising results

#* learning skewed class distributions
#@Maria Carolina Monard,Gustavo Batista
#t2002
#cAdvances in Logic, Artificial Intelligence and Robotics
#index223
#id223
#%
#! aspects influence performance achieved classifier created machine learning aspects related difference numbers examples belonging class difference large learning difficulties learn concept related minority class work discuss issues related learning skewed class distributions relationship cost sensitive learning

#* robust metric opinion
#@K Nigam,Matthew Hurst
#t2004
#cAAAI Spring Symposium on Exploring Attitude and Affect in Text
#index224
#id224
#%
#! paper describes automated detecting polar expressions topic elementary components approach shallow nlp polar language extraction machine learning based topic classifier components composed making simple accurate collocation assumption topical sentence contains polar language predicts polar language reflective topic subject matter evaluate components assumption corpus online consumer messages based components discuss measure overall sentiment particular topic expressed online messages authored different people propose use fundamentals bayesian statistics form aggregate authorial opinion metric metric propagate uncertainties introduced polarity topic modules facilitate statistically valid comparisons opinion multiple topics

#* bayesian approach filtering junk mail
#@Mehran Sahami,S. Dumais,D. Heckerman,Eric Horvitz
#t1998
#cWorkshop on Learning for Text Categorization
#index225
#id225
#%
#! addressing growing problem junk mail internet examine methods automated construction filters eliminate unwanted messages user mail stream casting problem decision theoretic framework able make use probabilistic learning methods conjunction notion differential misclassification cost produce filters especially appropriate nuances task appear

#* machine learning automated text categorization
#@Fabrizio Sebastiani
#t2002
#cACM Comput. Surv.
#index226
#id226
#%
#! automated categorization classification texts predefined categories witnessed booming years increased availability documents digital form ensuing need organize research community dominant approach problem based machine learning techniques general inductive process automatically builds classifier learning set preclassified documents characteristics categories advantages approach knowledge engineering approach consisting manual definition classifier domain experts good effectiveness considerable savings terms expert labor power straightforward portability different domains survey discusses main approaches text categorization fall machine learning paradigm discuss issues pertaining different problems document representation classifier construction classifier evaluation

#* makes weblog weblog
#@Dave Winer
#t2003
#cnull
#index227
#id227
#%
#! berkman studying weblogs used saying know wanted list known features weblog software important heart weblog weblog different wiki news site managed software like vignette interwoven draw experience developing using weblog software manila radio userland using competitive products blogger movable type piece published keynotes oscom jupiter weblogs conference disclaimer work progress subsequent versions art market weblog software develops

#* automated ranking database query results
#@Sanjay Agrawal,Surajit Chaudhuri,Gautam Das,Aristides Gionis
#t2003
#cCIDR
#index228
#id228
#%229,228,197,230,231,232,233,234,235,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250,251,252
#! ranking returning relevant results query popular paradigm information retrieval discuss challenges investigate approaches enable ranking databases including adaptations known techniques information retrieval present results preliminary experiments

#* dbxplorer keyword based search relational databases
#@Sanjay Agrawal
#t2002
#cICDE
#index229
#id229
#%
#! internet search engines popularized keywordbased search paradigm traditional database management systems offer powerful query languages allow keyword based search paper discuss dbxplorer enables keywordbased search relational databases dbxplorer implemented using commercial relational database web server allows users interact browser end outline challenges discuss implementation including results extensive experimental evaluation

#* empirical analysis predictive algorithms collaborative filtering
#@John S. Breese,D. Heckerman,Carl Myers Kadie
#t1998
#cUAI
#index230
#id230
#%
#! collaborative filtering recommender systemsuse database user preferences topredict additional topics products newuser likefi paper describeseveral algorithms designed task includingtechniques based correlation coefficients vector based similarity calculations statistical bayesian methodsfi comparethe predictive accuracy variousmethods set representative problemdomainsfi use basic classes evaluation

#* selection queries relational databases mapping strategies performance evaluation
#@Nicolas Bruno,Surajit Chaudhuri,Luis Gravano
#t2002
#cACM Trans. Database Syst.
#index231
#id231
#%
#! applications users specify target values certain attributes requiring exact matches values return instead result queries typically rank tuples best match given attribute values paper study advantages limitations processing query translating single range query traditional relational database management rdbms process efficiently particular study determine range query evaluate query exploiting statistics available rdbms impact quality statistics retrieval efficiency resulting scheme report experimental evaluation mapping strategies real rdbms microsoft sql server experiments new techniques robust significantly efficient previously known strategies requiring sequential scan data sets

#* evaluating queries web accessible databases
#@Amelie Marian,Nicolas Bruno,Luis Gravano
#t2004
#cACM Trans. Database Syst.
#index232
#id232
#%
#! query web search engine usually consists list keywords search engine responds best pages query query model prevalent multimedia collections general plain relational data certain applications example consider relation information available restaurants including location price range diner overall food rating user queries relation simply specify user location target price range expect return best restaurants terms combination proximity user closeness match target price range overall food rating processing queries efficiently challenging number reasons critical reason web applications relation attributes available external web accessible form interfaces query repeatedly potentially large set candidate objects paper study process topk queries efficiently setting attributes users specify target values handled external autonomous sources variety access interfaces present algorithms processing queries evaluate thoroughly using synthetic real web accessible data

#* saying sql
#@Michael J. Carey,Donald Kossmann
#t1997
#c1997	ACM  SIGMOD Record
#index233
#id233
#%
#! paper study simple sql extension enables query writers explicitly limit cardinality query result examine impact query optimization run time execution components relational dbms presenting approaches conservative approach aggressive approach exploiting cardinality limits relational query plans results obtained empirical study conducted using db demonstrate benefits sql extension illustrate tradeoffs approaches implementing

#* reducing braking distance sql query engine
#@Michael J. Carey,Donald Kossmann
#t1998
#cVLDB
#index234
#id234
#%
#! recent paper proposed adding stop clause tosql permit cardinality query result explicitly limitedby query writers query toolsfi demonstrated usefulnessof having clause showed extend traditionalcost based query optimizer accommodate demonstratedvia db based simulations large performance gains possiblewhen stop queries explicitly supported database engine paper present new

#* integration heterogeneous databases common domains using queries based textual similarity
#@William W. Cohen
#t1998
#cSIGMOD Conference
#index235
#id235
#%
#! databases contain constants like course numbers personal names place names correspond entities real worldfi previous work integration heterogeneous databases assumed local constants mapped appropriate global domain normalizationfi cases assumption does hold determining constants considered identical require detailed knowledge world purpose user query

#* providing database like access web using queries based textual similarity
#@William W. Cohen
#t1998
#cSIGMOD Conference
#index236
#id236
#%
#! databases contain constants like course numbers personal names place names correspond entities real world previous work integration heterogeneous databases assumed local constants mapped appropriate global domain normalization assume instead names given natural language text propose logic database integration called whirl reasons explicitly similarity local names measured using vector space model commonly adopted statistical information retrieval implemented data integration based whirl used successfully integrate information dozen web sites domains

#* fuzzy queries multimedia database systems
#@Ronald Fagin
#t1998
#cPODS
#index237
#id237
#%
#! essential differences multimedia databases containcomplicated objects images traditional database differenceslead interesting new issues particular cause consider newtypes queriesfi example multimedia database reasonable naturalto ask images similar fixed imagefi furthermore different ways obtaining accessing information multimediadatabase

#* optimal aggregation algorithms middleware
#@Ronald Fagin,Amnon Lotem,Moni Naor
#t2003
#cJ. Comput. Syst. Sci.
#index238
#id238
#%
#! assume object database grades scores eachof attributes example object color grade tells red shape grade tells round attribute sorted list lists object grade attribute sorted grade highest grade object assigned overall grade obtained combining attribute grades using fixed monotone aggregation function combining rule suchas min average determine objects objects highest overall grades naive algorithm access object database grade attribute fagin given algorithm fagin algorithm fa efficient monotone aggregation functions fa optimal high probability worst case analyze elegant remarkably simple algorithm threshold algorithm ta optimal stronger sense fa ta essentially optimal just monotone aggregation functions just high probability worst case sense database unlike fa requires large buffers size grow unboundedly database size grows ta requires small constant size buffer ta allows early stopping yields precise sense approximate version answers distinguish types access sorted access middleware obtains grade object sorted list proceeding list sequentially random access middleware requests grade object list obtains step consider scenarios random access impossible expensive relative sorted access provide algorithms essentially optimal cases elsevier science usa rights reserved

#* fastmap fast algorithm indexing data mining visualization traditional multimedia datasets
#@Christos Faloutsos,King-Ip (David) Lin
#t1995
#cSIGMOD Rec.
#index239
#id239
#%
#! promising idea fast searching traditional multimedia databases map objects points space using feature extraction functions provided domain expert subsequently use highly fine tuned spatial access methods sams answer types queries including query example type translates range query

#* probabilistic framework vague queries imprecise information databases
#@Norbert Fuhr
#t1990
#cVLDB
#index240
#id240
#%
#! probabilistic learning model vague queries missing imprecise information databases described instead retrieving set answers approach yields ranking objects database response query using relevance judgements user objects retrieved ranking actual query overall retrieval quality improved specifying different kinds conditions vague queries notion vague predicates introduced based underlying probabilistic model imprecise missing attribute values treated easily addition corresponding formulas applied combination standard predicates valued logic extending standard database systems coping missing imprecise data

#* probabilistic relational model integration ir databases
#@Norbert Fuhr
#t1993
#cSIGIR
#index241
#id241
#%
#! paper probabilistic relational model presented combines relational algebra probabilistic retrieval based certain independence assumptions operators relational algebra redefined probabilistic algebra generalization standard relational algebra furthermore special join operator implementing probabilistic retrieval proposed applied typical document databases queries ask documents kind object database addition implicit ranking objects provided case query relates probabilistic indexing uses probabilistic join operator proposed algebra intended standard interface combined database ir systems basis implementing user friendly interfaces

#* preference sql design implementation experiences
#@Werner Kiesling,Gerhard Kostler
#t2002
#cVLDB
#index242
#id242
#%
#! current search engines hardly cope ade quately fuzzy predicates defined complex preferencesfi biggest problem search engines implemented standard sql sql does directly understand notion preferencesfi preference sql extends sql preference model based strict partial orders presented companion paper preference queries behave like soft selection constraintsfi built base preference types powerful pareto operator combined adherence declarative sql programming style guarantees great programming producti vityfi preference sql optimizer does effi cient writing standard sql including high level implementation skyline opera tor pareto optimal setsfi pre processor approach enables seamless application

#* foundations preferences database systems
#@Werner Kiesling
#t2002
#cVLDB
#index243
#id243
#%
#! personalization services poses new challenges database technology demanding powerful flexible modeling technique complex preferences preference queries answered cooperatively treating preferences soft constraints attempting best possible match making propose strict partial order semantics preferences closely matches people intuition variety natural sophisticated preferences covered model inductively construct complex preferences means various preference constructors model key new discipline called preference engineering preference algebra given best matches bmo query model investigate complex preference queries decomposed simpler ones preparing ground divide conquer algorithms standard sql xpath extended seamlessly preferences presented companion paper believe model appropriate extend database technology effective support personalization

#* vague user interface relational databases permits vague queries
#@Amihai Motro
#t1988
#cACM Trans. Inf. Syst.
#index244
#id244
#%
#! specific query establishes rigid qualification concerned data match precisely vague query establishes target qualification concerned data close target conventional database systems handle vague queries directly forcing users retry specific queries repeatedly minor modifications match data satisfactory article describes called vague handle vague queries directly principal concept vague extension relational data model data metrics definitions distances values domain problem implementing data distances different users different interpretations notion distance vague incorporates features enable adapt itself individual views priorities users

#* experiences mining aviation safety data
#@Zohreh Nazeri,Eric Bloedorn,Paul Ostwald
#t2001
#cSIGMOD Conference
#index245
#id245
#%
#! goal data analysis aviation safety simple improve safety path goal hard identify data mining methods applicable task data available analyzed focus interesting results answers questions based recent research project completed encouraging news number aviation safety offices doing commendable work collect analyze safety related data number areas data mining techniques provide new tools perform analyses considered easily currently aviation safety offices collect analyze incident reports combination manual automated methods data analysis safety officers familiar domain data mining methods aviation safety officers tools automate database query report generation process actual analysis officer fairly rudimentary tools help extract useful information data research project looked application data mining techniques aviation safety data help aviation safety officers analysis task effort led creation tool called aviation safety data mining workbench paper describes research effort workbench experience data mining aviation safety data lessons learned

#* approach integrating query refinement sql
#@Michael Ortega-Binderberger,Kaushik Chakrabarti,Sharad Mehrotra
#t2002
#cEDBT
#index246
#id246
#%
#! emergence applications require content based similarity retrieval techniques support retrieval paradigm database systems emerged critical area researchfi user subjectivity important aspect queries ifiefi objects relevant user depends perception userfi query refinement used handle user subjectivity similarity search systemsfi paper explores enhance database systems

#* content based image retrieval relevance feedback mars
#@Yong Rui,T.Huang,Sharad Mehrotra
#t1997
#cICIP (2)
#index247
#id247
#%
#! technology advances areas image processing ip information retrieval ir evolved separately long time successful content based image retrieval systems require integration urgent need develop integration mechanisms link image retrieval model text retrieval model established text retrieval techniques utilized approaches converting image feature vectors domain weighted term vectors ir domain proposed paper furthermore relevance feedback technique ir domain used content based image retrieval demonstrate effectiveness conversion experimental results image retrieval precision increases considerably using proposed integration approach

#* density estimation
#@
#t1000
#cEMPTY
#index248
#id248
#%
#! articlefi introduce classic nonparametric estimator histogram outline theoretical properties good practicefi demonstrate toimprove histogram leading discussion popular kernel methodsfi concludewith bivariate example way choosing smoothing parameters new directionsthat promise improvementsfiwhy choose nonparametric parametric density estimation parametric densityestimation requires proper specification

#* index based xxl search engine querying xml data relevance ranking
#@Anja Theobald,Gerhard Weikum
#t2002
#cEDBT
#index249
#id249
#%
#! query languages xml xpath xquery support boolean retrieval query result possibly restructured subset xml elements entire documents satisfy search conditions query search paradigm works highly schematic xml data collections electronic catalogs searching information open environments web intranets large corporations ranked retrieval appropriate query result rank list xml

#* improved heterogeneous distance functions
#@D. Randall Wilson,Tony R. Martinez
#t1997
#cCoRR
#index250
#id250
#%
#! instance based learning techniques typically handle continuous linear input values handle nominal input attributes appropriately value difference metric vdm designed reasonable distance values nominal attribute values itlargely ignores continuous attributes requiring discretization map continuous values intonominal values paper proposes new heterogeneous distance functions called heterogeneous value difference metric hvdm interpolated value difference metric ivdm windowed value difference metric wvdm new distance functionsare designed handle applications nominal attributes continuous attributes inexperiments applications new distance metrics achieve higher classification accuracyon average previous distance functions datasets nominal andcontinuous attributes

#* using fagin algorithm merging ranked results multimedia middleware
#@Edward L. Wimmers,Laura M. Haas,Mary Tork Roth,Christoph Braendli
#t1999
#cCoopIS
#index251
#id251
#%
#! distributed multimedia information allows users access data different modalities different data sources ranked various combinations criteria fagin gives algorithm efficiently merging multiple ordered streams ranked results form new stream ordered combination ranks paper implementation fagin algorithm actual multimedia middleware including novel incremental version algorithm supports dynamic exploration data algorithm perform single multimedia server effective distributed environment limited set queries assumptions makes random access limit applicability dramatically experience provides better understanding important algorithm exposes open problem distributed multimedia information systems

#* falcon feedback adaptive loop content based retrieval
#@Leejay Wu,Christos Faloutsos,Katia P. Sycara,Terry R. Payne
#t2000
#cVLDB
#index252
#id252
#%
#! methods currently exist canperform relatively simple queries drivenby relevance feedback large multimediadatabases methods workonly vector spaces requirethat objects represented vectors withinfeature spaces implied queryregions typically convex research pa explains solution propose novel method designedto handle disjunctive queries metricspaces user provides weights pos itive examples learns im plied concept returns similar objects method diers existing relevance feedback methods base uponeuclidean mahalanobis metrics facili tates learning disjunctive concave mod els vector spaces arbitrarymetric spaces addition method iscompletely example driven imposes norequirements user aspectssuch feature selection main contributions fold onlydo present novel way estimate dissimilarity object set desir able objects support algo rithm shows exploit metric index ing structures support range queries toaccelerate search incurring falsedismissals empirical results demonstrate method converges rapidly excellentprecision recall outperforming sequen tial scanning

#* clustering documents words using bipartite spectral graph partitioning
#@Inderjit S. Dhillon
#t2001
#cKDD
#index253
#id253
#%254,255,256,257,258,259,260,261,262,97,170,263,264,120,265,266,267,268,269
#! document clustering word clustering studied problems existing algorithms cluster documents words separately simultaneously paper present novel idea modeling document collection bipartite graph documents words using simultaneous clustering problem posed bipartite graph partitioning problem solve partitioning problem use new spectral clustering algorithm uses second left right singular vectors appropriately scaled word document matrix yield good bipartitionings spectral algorithm enjoys optimality properties shown singular vectors solve real relaxation np complete graph bipartitioning problem present experimental results verify resulting clustering algorithm works practice

#* distributional clustering words text classification
#@L. Douglas Baker,A McCallum
#t1998
#cSIGIR
#index254
#id254
#%
#! paper applies distributional clustering pereiraet al document classication approachclusters words groups based distributionof class labels associated word like otherunsuperviseddimensionality reductiontechniques latent semantic indexing areable compress feature space aggres sively maintaining high document classi cation accuracy experimental results obtained real worlddata sets reduce feature dimen sionality orders magnitude lose accuracysignicantly better latent seman tic indexing deerwesteret al class based clus tering brownet al feature selection mutualinformation yang pederson markov blanket based feature selection koller sahami aggressive clusteringsometimes results improved classication accuracyover classication clustering

#* hierarchical taxonomies using divisive partitioning
#@Daniel Boley
#t1998
#cTechnical Report
#index255
#id255
#%
#! propose unsupervised divisive partitioning algorithm document data sets enjoys favorable propertiesfi particular algorithm shows excellent scalability large data collections produces high quality clusters competitive clustering methodsfi algorithm yields information significant distinctive words cluster words inserted naturally occuring hierarchical structure produced algorithmfi

#* document categorization query generation world wide web using webace
#@Daniel Boley,Maria L. Gini,Robert Gross,Eui-Hong Han,Kyle Hastings,G Karypis,Vipin Kumar,Bamshad Mobasher,Jerome Moore
#t1999
#cArtif. Intell. Rev.
#index256
#id256
#%
#! present webace agent exploring categorizing documents world wide web based user profilefi heart agent unsupervised categorization set documents combined process generating new queries used search new related documents filtering resulting documents extract ones closely related starting setfi document categories given priorifi present overall architecture

#* cluster based approach thesaurus construction
#@Carolyn J. Crouch
#t1988
#cSIGIR
#index257
#id257
#%
#! importance thesaurus successful operation information retrieval recognized techniques support automatic generation thesauri remain largely undiscovered paper describes approach automatic generation global thesauri based discrimination value model salton yang yu appropriate clustering algorithm method implemented applied document collections preliminary results indicate method produces improvements retrieval performance excess percent test collections viable worthy continued investigation

#* scatter gather cluster based approach browsing large document collections
#@Douglas R. Cutting,Jan O. Pedersen,David R. Karger,J. W. Tukey
#t1992
#cSIGIR
#index258
#id258
#%
#! document clustering received informationretrieval toolfi objections use fall intotwo main categories clustering slow forlarge corpora running time quadratic number documents second clustering doesnot appreciably improve retrievalfiwe argue problems arise clusteringis used attempt improve conventional searchtechniquesfi looking clustering informationaccess

#* efficient clustering large document collections
#@Inderjit S. Dhillon,James Fan,Yuqiang Guan
#t2001
#cData Mining for Scientific and Engineering Applications
#index259
#id259
#%
#! invaluable portion scientific data occurs naturally text formfi given large unlabeled document collection helpful organize collection clusters related documentsfi using vector space model text data treated high dimensional sparse numerical data vectorsfi contemporary challenge efficiently preprocess cluster large document collectionsfi paper present time memory ecient technique entire clustering

#* concept decompositions large sparse text data using clustering
#@Inderjit S. Dhillon,Dharmendra S. Modha
#t2001
#cMachine Learning
#index260
#id260
#%
#! unstructured text documents increasingly common available mining data sets represents major contemporary challengefi using words features text documents represented high dimensional sparse vectors thousand dimensions sparsity typicalfi paper study certain spherical means algorithm clustering document vectorsfi algorithm outputs disjoint clusters concept vector centroid

#* lower bounds partitioning graphs
#@W. E. Donath,A. J. Hoffman
#t1973
#cIBM Journal of Research and Development
#index261
#id261
#%
#! let partition graph division vertices disjoht subsets containing mk vertices let number edges vertices belong different subsets let largest eigenvalues matrix sum adjacency matrix graph plus diagonal matrix sum elements sum matrix zero theorem given shows effect maximum degree node limited shown right hand concave function computational studies ratio upper bound lower bound partition number random graphs having nodes

#* pattern classification
#@Richard O.Duda,Peter E.Hart,David G. Stork
#t2000
#cbook
#index262
#id262
#%
#! pattern classification consists assigning entities described featurevectors predefined groups patternsfi statistical characteristicsof problem consideration perfectly known minimalerror probability achieved means bayes decision rulefiin practice sub optimal classifier constructed fromtraining data neural network approaches problem havebeen proposedfi nearest neighbor models based assessing

#* study document representations multidimensional scaling indexing terms
#@Katter,V. Robert
#t1967
#cnull
#index263
#id263
#%
#! investigation continuing project better means studying evaluating document representations indexes present study focused understanding ways individuals interpret index terms goal using findings improving indexing systems indexing systems use bases matching descriptors documents descriptors information requests humans judging similarity meaning different descriptors respond terms degrees similarity appears discrimination power available human judgments matching arrangements question data human judgments similarity indexing terms available machine searching search accuracy improved present levels conclusions incorporation similarity relations mechanical index term matching procedures increase search accuracy procedures used study adapted use

#* self organizing maps
#@Tom Germano
#t1999
#cbook
#index264
#id264
#%
#! self organizing maps soms data visualization technique invented professor teuvo kohonen reduce dimensions data use self organizing neural networks problem data visualization attempts solve humans simply visualize high dimensional data techniques created help understand high dimensional data techniques reducing dimensions data presented course land multi dimensional scaling way soms reducing dimensions producing map usually dimensions plot similarities data grouping similar data items soms accomplish things reduce dimensions display similarities just idea som looks like example som constructed like colors grouped greens upper left hand corner purples grouped lower right right hand

#* projections efficient document clustering
#@Hinrich Schutze,Craig Silverstein
#t1997
#cSIGIR
#index265
#id265
#%
#! clustering increasing importance linear constant time clustering algorithms slow real time applicationsfi simple way speed clustering speed distance calculations heart clustering routinesfi study techniques improving cost distance calculations lsi truncation determine techniques speed clustering affect quality resulting clustersfi

#* normalized cuts image segmentation
#@Jianbo Shi,Jitendra Malik
#t2000
#cIEEE Trans. Pattern Anal. Mach. Intell.
#index266
#id266
#%
#! propose novel approach solving perfi ceptual grouping problem vision fofi cusing local features consistencies image data approach aims extracting global impression image treat image segmentafi tion graph partitioning problem propose novel global criterion normalized cut segmentfi ing graph normalized cut criterion measures total dissimilarity different groups total similarity groups cient computational technique based generalized eigenvalue problem used opfi timize criterion applied approach segmenting static images results enfi couraging

#* impact similarity measures web page clustering
#@Alexander Strehl,Er Strehl,J Ghosh,Raymond Mooney
#t2000
#cWorkshop on Artificial Intelligence for Web Search (AAAI 2000)
#index267
#id267
#%
#! clustering web documents enables semi automatedcategorization facilitates certain types searchfiany clustering method embed documentsin suitable similarity spacefi clusteringmethods associated similarity measures havebeen proposed past systematic comparativestudy impact similarity metrics oncluster quality possibly popular cost criteriado readily translate qualitatively different

#* information retrieval
#@
#t2008
#cEncyclopedia of GIS
#index268
#id268
#%
#! information retrieval wide loosely defined term pages shall concerned automatic information retrieval systemsfi automatic opposed manual information opposed data factfi unfortunately word information misleadingfi context information retrieval ir information technical meaning given shannon theory communication readily measured shannon weaver fi fact cases

#* effectiveness efficiency agglomerative hierarchic clustering document retrieval
#@EM Voorhees
#t1986
#cInformation Processing and Management
#index269
#id269
#%
#! ad hoc network cooperative engagement collection mobile hosts required intervention centralized access point paper present innovative design operation ad hoc networks basic idea design operate mobile host specialized router periodically advertises view interconnection topology mobile hosts network amounts new sort routing protocol investigated modifications basic bellman ford routing mechanisms specified rip make suitable dynamic self starting network mechanism required users wishing utilize ad hoc networks modifications address previous objections use bellman ford related poor looping properties algorithms face broken links resulting time dependent nature interconnection topology describing links mobile hosts finally ways basic network layer routing modified provide mac layer support ad hoc networks

#* combining language model sentiment analysis opinion
#@Xiangwen Liao,Donglin Cao,Songbo Tan,Yue Liu,Guodong Ding
#t2006
#cOnline Proceedings of Text Retrieval Conference (TREC)
#index270
#id270
#%271,272,273
#! paper describes participation blog opinion retrieval task year conduct experiments firtex platform developed lab language model used retrieve related blog unit interactive knowledge adopted expand query retrieve blog unit include opinion introduce novel extracting technology extract text retrieved blog post finally lexicon based method used rerank document opinion

#* language modeling approach information retrieval
#@Jay M. Ponte,W. Bruce Croft
#t1998
#cSIGIR
#index271
#id271
#%
#! models document indexing document retrieval extensively studied integration classes models goal researchers difficult problem argue reason lack adequate indexing model suggests better indexing model help solve problem feel making unwarranted parametric assumptions lead better retrieval performance furthermore making prior assumptions similarity documents warranted instead propose approach retrieval based probabilistic language modeling estimate models document individually approach modeling non parametric integrates document indexing document retrieval single model advantage approach collection statistics used heuristically retrieval models integral model implemented model tested empirically approach significantly outperforms standard tf idf weighting different collections query sets

#* automatic query expansion using smart
#@Chris Buckley,Gerard Salton,James Allan,Amit Singhal
#t1995
#cProceedings of The third Text REtrieval Conference (TREC-3)
#index272
#id272
#%
#! smart information retrieval project emphasizes completely automatic approaches understanding retrieval large quantities text continue work trec performing runs routing ad hoc foreign language environments major focus massive query expansion adding terms query terms come known relevant documents case routing just retrieved documents case ad hoc spanish

#* hierarchical dirichlet language model natural language engineering
#@D. J. C. Mackay,L. Peto
#t1994
#cNatural Language Engineering
#index273
#id273
#%
#! discuss hierarchical probabilistic model predictions similar popular language modelling procedure known smoothing number interesting differences smoothing emerge insights gained probabilistic view problem point new directions language modelling ideas paper applicable problems modelling triphomes speech dna protein sequences molecular biology

#* constrained means clustering
#@P. S. Bradley,K. P. Bennett,A. Demiriz
#t2000
#cMSR-TR-2000-65, Microsoft Research
#index274
#id274
#%275,276,277
#! consider practical methods adding constraints means clustering algorithm order avoid local solutions clusters clusters having points observe phenomena applying means datasets number dimensions number desired clusters propose explicitly adding constraints underlying clustering optimization problem requiring cluster minimum number points investigate resulting cluster assignment step preliminary numerical tests real datasets indicate constrained approach prone poor local solutions producing better summary underlying data contrained means clustering

#* density based indexing approximate nearest neighbor queries
#@K. P. Bennett,Usama Fayyad,Dan Geiger
#t1999
#cACM SIGKDD Conference Proceedings
#index275
#id275
#%
#! consider problem performing nearest neighbor queries efficiently large high dimensional databases assuming database scan determine nearest neighbor entries acceptable study possibility constructing index structure database accepted traditional database indexing algorithms fail high dimensional data say depending scheme arguments haveadvocated nearest neighbor queries

#* nonlinear programming
#@D.P. Bertsekas 1999
#t0
#cnull
#index276
#id276
#%
#! substantially expanded pages improved edition best selling nonlinear programming book treatment focuses iterative algorithms constrained unconstrained optimization lagrange multipliers duality large scale problems interface continuous discrete optimization

#* compressed data cubes olap aggregate query approximation continuous dimensions
#@Jayavel Shanmugasundaram,Usama Fayyad,P. S. Bradley
#t1999
#cKDD
#index277
#id277
#%
#! efficiently answering decision support queries importantproblem work direction contextof data cube queries efficiently answered pre computinglarge parts cube having large space requirements pre computation requires hierarchy eachdimension fixed dimensions categorical pre discretized queries advantage pre computation canthus drill roll fixed hierarchy anotherdisadvantage existing pre computation techniques thetarget measure aggregation function isfixed cube queries target measure orusing different aggregation functions require pre computinglarger data cubes paper propose new compressedrepresentation data cube drastically reduces storagerequirements does require discretization hierarchy alongeach query dimension fixed treats eachdimension potential target measure supports multipleaggregation functions additional storage costs tradeoffis approximate relatively accurate answers queries weoutline mechanisms reduce error approximation ourperformance evaluation indicates compression techniqueeffectively addresses limitations existing approaches

#* constraint driven clustering
#@Rong Ge,Martin Ester,Wen Jin,Ian Davidson
#t2007
#cKDD
#index278
#id278
#%0,279,280,281,282,283,274,284,285,286,287,288,156,289,290,291,292,293,159,294,295,296,297,298,299,300,301,302,303
#! clustering methods data driven need driven data driven methods intend discover true structure underlying data need driven methods aims organizing true structure meet certain application requirements need driven constrained clustering able useful actionable clusters applications energy aware sensor networks privacy preservation market segmentation existing methods constrained clustering require users provide number clusters unknown advance crucial impact clustering result paper argue natural way generate actionable clusters let application specific constraints decide number clusters purpose introduce novel cluster model constraint driven clustering cdc finds priori unspecified number compact clusters satisfy user provided constraints general types constraints considered minimum significance constraints minimum variance constraints combinations types prove np hardness cdc problem different constraints propose novel dynamic data structure cd tree organizes data points leaf nodes leaf node approximately satisfies cdc constraints minimizes objective function based cd trees develop efficient algorithm solve new clustering problem experimental evaluation synthetic real datasets demonstrates quality generated clusters scalability algorithm categories subject descriptors

#* approximation algorithms anonymity
#@Gagan Aggarwal,Tomas Feder,Krishnaram Kenthapadi,Rajeev Motwani,Rina Panigrahy,Dilys Thomas,An Zhu
#t2005
#cJournal of Privacy Technology
#index279
#id279
#%
#! consider problem releasing table containing personal records ensuring individual privacy maintaining data integrity extent possible techniques proposed literature anonymization release considered anonymous information corresponding individual release distinguished individuals information appears release order achieve anonymization entries table suppressed generalized age value changed age range goal lose little information possible ensuring release anonymous optimization problem referred anonymity problem anonymity problem np hard attribute values ternary allowed suppress entries positive provide approximation algorithm problem improved positive results interesting cases specific values particular approximation algorithm special case anonymity approximation algorithm anonymity

#* max min cluster formation wireless ad hoc networks
#@Alan D. Amis,Ravi Prakash,Dung Huynh,Thai Vuong
#t2000
#cINFOCOM
#index280
#id280
#%304,305,306,307,308,309,310,311,312,313,123,314
#! ad hoc network logically represented set clusters clusterheads form hop dominating set node hops clusterhead clusterheads form virtual backbone used route packets nodes cluster previous heuristics restricted hop clusters minimum hop dominating set problem np complete present heuristic form clusters wireless ad hoc network nodes assumed non deterministic mobility pattern clusters formed diffusing node identities wireless links heuristic terminates node clusterhead wireless hops away clusterhead value parameter heuristic heuristic run regular intervals network configuration changes features heuristic tends elect existing clusterheads network configuration changes helps reduce communication overheads transition old clusterheads new clusterheads tendency evenly distribute mobile nodes clusterheads evently distribute responsibility acting clusterheads nodes heuristic fair stable simulation experiments demonstrate proposed heuristic better earlier heuristics lca degree based solutions

#* energy efficient hierarchical clustering algorithm wireless sensor networks
#@Seema Bandyopadhyay,Edward J. Coyle
#t2003
#cINFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE Computer and Communications. IEEE Societies
#index281
#id281
#%
#! wireless network consisting large number small sensors low power transceivers effective tool gathering data variety environments data collected sensor communicated network single processing center uses reported data determine characteristics environment detect event communication message passing process designed conserve limited energy resources sensors clustering sensors groups sensors communicate information clusterheads clusterheads communicate aggregated information processing center save energy paper propose distributed randomized clustering algorithm organize sensors wireless sensor network clusters extend algorithm generate hierarchy clusterheads observe energy savings increase number levels hierarchy results stochastic geometry used derive solutions values parameters algorithm minimize total energy spent network sensors report data clusterheads processing center

#* scalable clustering algorithms balancing constraints
#@A Banerjee,J Ghosh
#t2006
#cData Min. Knowl. Discov.
#index282
#id282
#%
#! clustering methods data mining problems extremely scalable addition data mining applications demand clusters obtained balanced approximately size importance paper propose general framework scalable balanced clustering data clustering process broken steps sampling small representative subset points clustering sampled data populating initial clusters remaining data followed refinements simple uniform sampling original data sufficient representative subset high probability proposed framework allows large class algorithms used clustering sampled set focus popular parametric algorithms ease exposition present algorithms populate refine clusters algorithm populating clusters based generalization stable marriage problem refinement algorithm constrained iterative relocation scheme complexity overall method kn log obtaining balanced clusters data points compares favorably existing techniques balanced clustering addition providing balancing guarantees clustering performance obtained using proposed framework comparable better corresponding unconstrained solution experimental results datasets including high dimensional ones provided demonstrate efficacy proposed framework

#* clustering scheme hierarchical control multi hop wireless networks
#@Suman Banerjee,Samir Khuller
#t2001
#cINFOCOM
#index283
#id283
#%
#! paper present clustering scheme create hierarchical control structure multi hop wireless networks cluster defined subset vertices induced graph connected addition cluster required obey certain constraints useful management scalability hierarchy constraints met simultaneously general graphs clustering obtained wireless network topologies finally present efficient distributed implementation clustering algorithm set wireless nodes create set desired clusters

#* localized minimum energy broadcasting ad hoc networks
#@Julien Cartigny,David Simplot,Ivan Stojmenovic
#t2003
#cINFOCOM
#index284
#id284
#%
#! minimum energy broadcasting problem node adjust transmission power order minimize total energy consumption enable message originated source node reach nodes ad hoc wireless networkfi existing solutions node requires global network information including distances neighboring nodes network order decide transmission radiusfi paper new localized protocol

#* clustering constraints feasibility issues means algorithm
#@Ian Davidson,S. S. Ravi
#t2005
#cSDM
#index285
#id285
#%
#! recent work looked extending means algorithmto incorporate background information theform instance level link link constraintsfiwe introduce ways specifying additionalbackground information form constraintsthat operate instances beinterpreted conjunctions disjunctions instancelevel constraints easy implementfi wepresent complexity results feasibility clusteringunder

#* identifying generating easy sets constraints clustering
#@Ian Davidson,S. S. Ravi
#t2006
#cAAAI
#index286
#id286
#%
#! clustering constraints recent innovation ar tificial intelligence community yielded significantpractical benefit recent work shown negative forms constraints associated sub problem just finding feasible clustering np complete worst case results entire problem class say noth ing prevalent easy problem instances work large pockets theseproblem classes clustering constraints easyand using easy sets constraints yields better empiricalresults illustrate sufficient conditions fromgraph theory identify apriori easy problem stances present algorithms create large easy tosatisfy constraint sets

#* complexity non hierarchical clustering constraints
#@
#t1000
#cEMPTY
#index287
#id287
#%
#! recent work looked extending clustering algorithms instance level link ml link cl background information work introduces cluster level constraints influence inter cluster distances cluster composition addition background information useful providing better clustering results raises important feasibility question given collection constraints set data does exist partition data set satisfying constraints study complexity feasibility problem constraints separately combinations constraints results clearly delineate combinations constraints feasibility problem computationally intractable np complete problem efficiently solvable computational class consider ml cl constraints conjunctive disjunctive normal forms cnf dnf respectively ml constraints feasibility problem intractable cnf efficiently solvable dnf unfortunately cl constraints feasibility problem intractable cnf dnf effectively means cl constraints non trivial form efficiently incorporated clustering algorithms overcome introduce notion choice set constraints prove feasibility problem choice sets efficiently solvable ml cl constraints present empirical results indicate feasibility problem occurs extensively real world problems

#* practical data oriented microaggregation statistical disclosure control
#@Josep Domingo-Ferrer,Josep Maria Mateo-Sanz
#t2002
#cIEEE Trans. Knowl. Data Eng.
#index288
#id288
#%
#! microaggregation statistical disclosure control technique microdata disseminated statistical databases raw microdata individual records data vectors grouped small aggregates prior publication aggregate contain data vectors prevent disclosure individual information constant value preset data protector exact polynomial algorithms known date microaggregate optimally minimal variability loss methods literature rank data partition groups fixed size multivariate case ranking performed projecting data vectors single axis paper candidate optimal solutions multivariate univariate microaggregation problems characterized univariate case heuristics based hierarchical clustering genetic algorithms introduced data oriented try preserve natural data aggregates multivariate case fixed size hierarchical clustering microaggregation algorithms presented require data projected single dimension methods clearly reduce variability loss compared conventional multivariate microaggregation projected data

#* disc based approach data summarization privacy preservation
#@Rong Ge,Martin Ester,Wen Jin,Zengjian Hu
#t2006
#cSSDBM
#index289
#id289
#%
#! data summarization recognized fundamental operation database systems data mining important applications data compression privacy preservation existing methods cfvalues databubbles perform reasonably provide guarantees quality results paper introduce summarization approach numerical data based discs formalizing notion quality objective minimal set discs spheres satisfying radius significance constraint covering given dataset proposed problem np complete design different approximation algorithms algorithms quality guarantee scale large databases machinery approximation algorithms allows precise characterization heuristic algorithm heuristic efficient algorithm exploits multi dimensional index structures integrated database systems experiments heuristic algorithm generates summaries outperform state art data bubbles terms internal measures terms external measures using data summaries input clustering methods

#* optimal energy aware clustering sensor network
#@Soheil Ghiasi,Ankur Srivastava,Xiaojian Yang,Majid Sarrafzadeh
#t2002
#cSensors
#index290
#id290
#%
#! sensor networks fastest growing technologies potential changing lives drastically collaborative dynamic distributed computing communicating systems self organizing capabilities distributing task efficient computation challenges implementation systems energy dissipation clustering order maintain certain degree service quality reasonable lifetime energy needs optimized stage operation sensor node clustering important optimization problem nodes clustered easily able communicate considering energy optimization parameter clustering imperative paper study theoretical aspects clustering problem sensor networks application energy optimization illustrate optimal algorithm clustering sensor nodes cluster master balanced total distance sensor nodes master nodes minimized balancing clusters needed evenly distributing load master nodes minimizing total distance helps reducing communication overhead energy dissipation problem balanced clustering modeled mincost flow problem solved optimally using existing techniques

#* clustering visualization retail market baskets
#@J Ghosh,Alexander Strehl
#t2005
#cbook
#index291
#id291
#%
#! transaction analysis including clustering market baskets key application data mining retail industry domain specific requirements need obtaining easily interpretable actionable results exhibits challenging characteristics stemming fact data thousands features highly non gaussian sparse chapter proposes relationship based approach clustering data tries sidestep curse dimensionality issue working suitable similarity space instead original high dimensional feature space intermediary similarity space suitably tailored satisfy business criteria requiring customer clusters represent comparable amounts revenue apply efficient scalable graph partitioning based clustering techniques space output clustering algorithm used reorder data points resulting permuted similarity matrix readily visualized dimensions clusters showing bands visualization helpful assessing improving clustering example actionable recommendations splitting merging clusters easily derived guides user suitable number clusters results presented real retail industry data set thousand customers products

#* joint optimization customer segmentation marketing policy maximize long term profitability
#@Jedid-Jah Jonker,Nanda Piersma,Dirk Van den Poel
#t2004
#cExpert Syst. Appl.
#index292
#id292
#%
#! advent marketing media targeted direct mail internet marketing opportunities develop targeted marketing campaigns enhanced way organizationally economically feasible profitably support substantially larger number marketing segments problem segments distinguish actions different segments increases substantially environment systematic analytic procedure optimizing steps welcome study present joint optimization approach addressing issues segmentation customers homogeneous groups customers determining optimal policy action set available actions segment implement joint optimization framework direct mail setting charitable organization previous studies area highlighted importance following variables ecency requency onetary value use variables segment customers second step determine marketing policy optimal using markov decision processes following similar previous applications attractiveness stochastic dynamic programming procedure based long run maximization expected average profit contribution lies combination steps optimization framework obtain optimal allocation marketing expenditures control segment stability policy performance bootstrap procedure framework illustrated real life application results proposed model outperforms chaid segmentation

#* robust effective anonymity large databases
#@Wen Jin,Rong Ge,Weining Qian
#t2006
#cPAKDD
#index293
#id293
#%
#! challenge privacy preserving data mining lies respecting privacy requirements discovering original interesting patterns structures existing methods loose correlations attributes transforming different attributes independently guarantee minimum ion level required legal policies paper propose novel privacy preserving transformation framework distance based mining operations based concept privacy preserving microclusters satisfy privacy constraint significance constraint framework extends robustness state art anonymity model introducing privacy constraint minimum radius keeping effectiveness significance constraint minimum number corresponding data records privacy preserving microclusters public data mining purposes original data records kept private present efficient methods generating maintaining privacy preserving microclusters data mining operations clustering easily adapted public data represented microclusters instead private data records experiment demonstrates proposed methods achieve accurate clusterings results preserving privacy

#* efficient clustering algorithms self organizing wireless sensor networks
#@Rajesh Krishnan,David Starobinski
#t2006
#cAd Hoc Networks
#index294
#id294
#%
#! self organization wireless sensor networks involves network decomposition connected clusters challenging task limited bandwidth energy resources available networksfi paper make contributions improving efficiency self organization wireless sensor networksfi present novel approach message efficient clustering nodes allocate local growth budgets neighborsfi introduce algorithms

#* complexity optimal anonymity
#@Adam Meyerson,Ryan Williams
#t2004
#cPODS
#index295
#id295
#%
#! technique anonymization proposed literature alternative way release public formation ensuring data privacy data integrityfi prove general versions opti mal anonymization relations np hard includ ing suppression version amounts choosing minimum number entries delete relationfi present polynomial time algorithm optimal anonymity achieves approximation ratio inde pendent size database constantfi particular log approximation constant big fihowever runtime algorithm exponential kfi slightly clever algorithm removes condition log approximation degree relationfi believe algorithm potentially quite fast practicefi

#* generalizing data provide anonymity disclosing information
#@Pierangela Samarati,Latanya Sweeney
#t1998
#cACM SIGACT-SIGMOD-SIGART symposium on Principles of database systems
#index296
#id296
#%
#! proliferation information internet access fast computers large storage capacities increased volume information collected disseminated individuals existence data sources makes easier identify individuals private information released data believed anonymous time increasing demands organizations release individualized data aggregate statistical information explicit identifiers phone number removed encrypted releasing individualized data characteristic data term quasi identifiers exist allow data recipient identify individuals data refer paper provide computational disclosure technique releasing information private table identity individual released data refer definitively recognized approach protects linking data based concepts generalization stored values replaced semantically consistent precise alternatives anonymity table said provide anonymity contained data allow recipient associate released information set individuals smaller introduce notions generalized table minimal generalization table respect anonymity requirement optimization problem objective minimally distort data providing adequate protection algorithm given table efficiently computes preferred minimal generalization provide anonymity

#* scalable approach balanced high dimensional clustering market baskets
#@Alexander Strehl,J Ghosh
#t2000
#cHiPC
#index297
#id297
#%
#! paper presents opossum novel similarity based clustering approach based constrained weighted graph partitioningfi opossum particularly attuned real life market baskets characterized high dimensional highly sparse customer product matrices positive ordinal attribute values significant outliersfi built metis known highly efficient graph partitioning algorithm inherits scalable easily parallelizeable

#* anonymity model protecting privacy
#@Latanya Sweeney
#t2002
#cInternational Journal of Uncertainty, Fuzziness and Knowledge-Based Systems
#index298
#id298
#%
#! consider data holder hospital bank privately held collection person specific field structured data suppose data holder wants share version data researchersfi data holder release version private data scientific guarantees individuals subjects data identified data remain practically useful solution provided paper includes formal protection model named anonymity set accompanying policies deploymentfi release provides anonymity protection information person contained release distinguished individuals information appears releasefi paper examines identification attacks realized releases adhere anonymity unless accompanying policies respectedfi anonymity protection model important forms basis real world systems known datafly argus similar provide guarantees privacy protectionfi

#* constraint based clustering large databases
#@Anthony K. H. Tung,Raymond T. Ng,Laks V. S. Lakshmanan,Jiawei Han
#t2001
#cICDT
#index299
#id299
#%
#! constrained clustering finding clusters satisfy user specified constraints highly desirable applications paper introduce constrained clustering problem traditional clustering algorithms means handle scalable constraint clustering algorithm developed study starts finding initial solution satisfies user specified constraints refines solution performing confined object movements constraints algorithm consists phases pivot movement deadlock resolution phases finding optimal solution np hard propose heuristics algorithm scale large data sets using heuristic micro cluster sharing experiments effectiveness efficiency heuristics

#* clustering instance level constraints
#@Kiri Wagstaff,Claire Cardie
#t2000
#cAAAI/IAAI
#index300
#id300
#%
#! clustering algorithms conduct search space possible organizations data set paper propose types instance level clustering constraints link link constraints incorporated clustering algorithm aid search data sets tested results indicate incorporation surprisingly constraints increase clustering accuracy decreasing runtime investigate relative ects type constraint type contributes accuracy improvements depends behavior clustering algorithm constraints

#* drawing planar graphs
#@Rene Weiskircher
#t1999
#cDrawing Graphs
#index301
#id301
#%
#! book presents important fundamental theorems algorithms planar graph drawing easy understand constructive proofs extensively illustrated exercises included end chapter suitable use advanced undergraduate graduate level courses algorithms graph theory graph drawing information visualization computational geometry book serve useful reference source researchers field graph drawing software developers information visualization vlsi design cad

#* birch efficient data clustering method large databases
#@Tian Zhang,Raghu Ramakrishnan,Miron Livny
#t1996
#cSIGMOD Conference
#index302
#id302
#%
#! finding useful patterns large datasets attractedconsiderable recently widelystudied problems area identification clusters densely populated regions multi dimensional datasetfiprior work does adequately address problem largedatasets minimization costsfithis paper presents data clustering method namedbirch balanced iterative reducing clustering usinghierarchies demonstrates

#* scalable balanced model based clustering
#@Shi Zhong,J Ghosh
#t2003
#cSDM
#index303
#id303
#%315,316,317,318,274,253,319,320,321,322,323,324,91,325,326,327,328,329,330,331,297,332,333,267,334,302,335,336,337
#! paper presents general framework adapting generative model based clustering algorithm provide balanced solutions clusters comparable sizes partitional model based clustering algorithms viewed iterative step optimization process iterative model estimation sample assignment instead maximum likelihood ml assignment balanceconstrained approach used sample assignment step efficient iterative bipartitioning heuristic developed reduce computational complexity step make balanced sample assignment algorithm scalable large datasets demonstrate superiority approach regular ml clustering complex data arbitraryshape spatial data high dimensional text documents eeg time series keywords model based clustering scalable algorithms balanced clustering constrained clustering

#* architectural organization mobile radio network distributed algorithm
#@D.J. Baker,A. Ephremides
#t1981
#cIEEE Trans. Commun
#index304
#id304
#%
#! paper consider problem organizing set mobile radio equipped nodes connected network require reliable structure acquired maintained face arbitrary topological changes node motion failure require structure achieved use central controller propose develop self starting distributed algorithm establishes maintains connected architecture algorithm especially suited needs hf intra task force itf communication network discussed paper

#* better heuristic orthogonal graph drawing
#@T Biedl,G Kant
#t1998
#cComputational Geometry
#index305
#id305
#%
#! orthogonal drawing graph embedding plane edges drawn sequences horizontal vertical segments present linear time space algorithm draw connected graph orthogonally grid size theta bends edge bent twice particular non planar non biconnected planar graphs big improvement algorithm simple easy implement handles planar non planar

#* routing ad hoc networks using minimum connected dominating sets
#@Bevan Das,Vaduvur Bharghavan
#t1997
#cICC (1)
#index306
#id306
#%
#! impose virtual backbone structure ad hoc network order support unicast multicast fault tolerant routing ad hoc network virtual backbone differs wired backbone cellular networks key ways change nodes used primarily routing packets flows computing updating routes primary routes packets flows computed shortest paths computation virtual backbone necessary provide backup routes handle interim failures dynamic nature virtual backbone approach splits routing problem levels update virtual backbone update routes key contribution paper alternatives finding updating virtual backbone virtual backbone small possible use approximation minimum connected dominating set mcds ad hoc network topology virtual backbone hosts mcds maintain local copies global topology network shortest paths pairs nodes

#* design concept reliable mobile radio networks frequency hopping signaling
#@A. Ephremides,Anthony
#t1000
#cEMPTY
#index307
#id307
#%
#! new architecture mobile radio networks called linked cluster architecture described methods implementing architecture using distributed control techniques presented illustrate fully distributed control methods combined hierarchical control create network robust respect node loss connectivity changes distributed algorithms presented deal formation linkage clusters activation network links study performance network structuring algorithms simulation model developed use simulation construct software simulation tools illustrated simulation results shown example high frequency fif intratask force hf communication network

#* maca bi maca invitation
#@
#t1000
#cEMPTY
#index308
#id308
#%
#! paper introduces new wireless mac protocol maca bi maca invitation protocol simplified version known maca multiple access collision avoidance based request send clear send rts cts handshake clear send cts control message retained request send rts rts cts handshake suppressed maca bi preserving data collision free property robust maca problems protocol failures

#* distributed algorithms generating loopfree routes networks frequently changing topology
#@Eli M. Gafni,Student Member,Dimitri,P. Bertsekas,Senior Member
#t1981
#cIEEE Transactions on Communications
#index309
#id309
#%
#! consider problem maintaining communication nodes data network central station presence frequent topological changes example mobile packet radio networks argue flooding schemes significant drawbacks networks propose ag eneral class distributed algorithms establishing new loop free routes station node left route china nthge sn etwork topology virtue built redundancy algorithms typically activatedv ery infrequently evenw hen involve communication portoiof nth network heen materially affected topological change

#* multicluster mobile multimedia radio network
#@Mario Gerla,Jack Tzu-Chieh Tsai
#t1995
#cWireless Networks
#index310
#id310
#%
#! paper present multi cluster multi hop packet radio network architecturethat addresses challenges implements required featuresfiin mobile multi channel code multi hop environment topology isdynamically reconfigured handle mobilityfi routing bandwidth assignmentare designed meet various types traffic requirementsfi node clustering vc setup channel access control underlying features whichsupport architecture

#* approximation algorithms combinatorial problems
#@David S. Johnson
#t1974
#cJ. Comput. Syst. Sci.
#index311
#id311
#%
#! simple polynomial time heuristic algorithms finding approximate solutions various polynomial complete optimization problems analyzed respect worst case behavior measured ratio worst solution value chosen algorithm optimal value certain problems simple form knapsack problem optimization problem based satisfiability testing algorithms ratio bounded constant independent problem size number set covering problems simple algorithms yield worst case ratios grow log problem size problem finding maximum clique graph algorithm ratio does grow fast egr problem size egr depends algorithm

#* spatial reuse multihop packet radio networks
#@L. Kleinrock, J. Silvester 
#t1987
#cProceedings of the IEEE
#index312
#id312
#%
#! multihop packet radio networks present challenging problems network analyst designer communication channel shared network users critical resource order make efficient use shared resource variety channel access protocols promote organized sharing investigated sharing occur domains frequency time space paper concerned sharing channel reuse spatial domain survey results approaches topological design associated channel access protocols attempt optimize performance spatial reuse communication channel presented

#* selecting routers ad hoc wireless networks
#@AK Parekh
#t1994
#cProceedings SBT/IEEE Intl Telecommunications
#index313
#id313
#%
#! advent portable lap tops personal digital assistants increas ingly important enable communication large networks mobile usersfi cellular framework long way meeting need clear necessary support adfihoc architectures fixed base stations presentfi routing ad hoc network supported client devices network ifiefi act routersfi essential range transmitters quite limited ir links supporting mbffis fi ieee fi commit tee wireless local area networking recognized ad hoc architectures important component standardization processfi paper model adhoc network undirected graphfi nodes graph processors communicate incident edges broadcastfi processors know size network start topological informationfi goal select small subset nodes graph act routers ad hoc networkfi clearly set routers node adjacent node subsetfi subset graph theoretic terminology called dominating setfi finding smallest dominating set graph known np hardfi present fast distributed algorithm finds dominating set size network nodes links provably close minimum cardinality setfi

#* universality considerations vlsi circuits
#@Leslie G. Valiant
#t1981
#cIEEE Trans. Computers
#index314
#id314
#%
#! problem embedding interconnection pattern circuit dimensional surface minimal area discussed natural patterns graphs containing connections require omega super area order achieve compact embeddings restricted classes graphs considered example arbitrary trees bounded degree embedded linear area edges crossing planar graphs embedded efficiently crossovers allowed embedding

#* competitive learning algorithms vector quantization
#@Stanley C. Ahalt,Ashok K. Krishnamurthy,Prakoon Chen,Douglas E. Melton
#t1990
#cNeural Networks
#index315
#id315
#%
#! efficient representation encoding signals limited resources finite storage capacity restricted transmission bandwidth fundamental problem technical biological information processing systems typically realistic circumstances encoding communication messages deal different sources noise disturbances paper propose unifying approach data compression robust vector quantization explicitly deals channel noise bandwidth limitations random elimination prototypes resulting algorithm able limit detrimental ect noise general communication scenario addition presented model allows derive novel competitive neural networks algorithm covers topology preserving feature maps called neural gas algorithm maximum entropy soft max rule special cases furthermore continuation methods based noise models improve codebook design reducing sensitivity local minima exemplary application novel robust vector quantization algorithm image compression teleconferencing

#* frequency sensitive competitive learning clustering high dimensional hyperspheres
#@A Banerjee,J Ghosh
#t2002
#cNeural Networks
#index316
#id316
#%
#! paper derives competitive learning mechanisms principles obtain clusters comparable sizes inputs representatives normalized mechanisms effective achieving balanced grouping inputs high dimensional spaces illustrated experimental results clustering popular text data sets dimensional spaces respectively

#* model based gaussian non gaussian clustering
#@Jeffrey D. Banfield, Adrian E. Raftery 
#t1993
#cBiometrics
#index317
#id317
#%
#! classification maximum likelihood approach sufficiently general encompass current clustering algorithms including based sum squares criterion criterion friedman rubin currently implemented does allow specification features orientation size shape common clusters differ clusters restricted gaussian distributions does allow noise propose ways overcoming limitations reparameterization covariance matrix allows specify features clusters practical framework non gaussian clustering outlined means incorporating noise form poisson process described approximate bayesian method choosing number clusters given performance proposed methods studied simulation encouraging results methods applied analysis data set arising study diabetes results better previous analyses keywords include bayes factors classification diabetes hierarchical agglomeration iterative relocation mixture models

#* gentle tutorial em algorithm application parameter estimation gaussian mixture hidden markov models
#@J. Bilmes
#t1997
#cnull
#index318
#id318
#%
#! maximum likelihood parameter estimation problem expectationmaximization em algorithm used solutionfi form em algorithm given literaturefi develop em parameterestimation procedure applications finding parameters mixture ofgaussian densities finding parameters hidden markov model hmm ifiefi baum welch algorithm discrete

#* cure efficient clustering algorithm large databases
#@Sudipto Guha,Rajeev Rastogi,Kyuseok Shim
#t2001
#cInf. Syst.
#index319
#id319
#%
#! clustering data mining useful discovering groups identifying interesting distributions underlying data traditional clustering algorithms favor clusters spherical shapes similar sizes frag ile presence outliers propose new cluster ing algorithm called cure robust outliers identifies clusters having non spherical shapes wide variances size cure achieves representing cluster certain fixed number points gen erated selecting scattered points cluster shrinking center cluster specified fraction having representa tive point cluster allows cure adjust geometry non spherical shapes shrinking helps dampen effects outliers handle large databases cure employs combination random sampling par titioning random sample drawn data set partitioned partition partially clustered partial clusters clustered second pass yield desired clusters experimental results confirm quality clusters produced cure better existing algorithms furthermore demonstrate random sampling partitioning enable cure outperform existing algorithms scale large databases sacrificing cluster ing quality

#* sublinear time approximation scheme clustering metric spaces
#@Piotr Indyk
#t1999
#cMetric Spaces, Proc. 40th IEEE FOCS
#index320
#id320
#%
#! metric clustering problem defined follows given metric partition sets order minimize value math math math si paper approximation scheme problem

#* generalized clustering supervised learning data assignment
#@Annaka Kalton,Pat Langley,Kiri Wagstaff,Jungsoon P. Yoo
#t2001
#cKDD
#index321
#id321
#%
#! clustering algorithms increasingly importantin handling analyzing data considerable work beendone devising effective increasingly specific clusteringalgorithmsfi contrast developed generalizedframework accommodates diverse clustering algorithmsin systematic wayfi framework views clustering ageneral process iterative optimization includes modulesfor supervised learning instance assignmentfi theframework suggested

#* clusterings good bad spectral
#@Ravi Kannan,Santosh Vempala,Adrian Vetta
#t2000
#c Foundations of Computer Science, 2000. Proceedings. 41st Annual Symposium
#index322
#id322
#%
#! motivate develop natural bicriteria measure assessing quality clustering avoids drawbacks existing measures simple recursive heuristic shown polylogarithmicworstcase guarantees newmeasure main result article analysis popular spectral algorithm variant spectral clustering turns effectiveworst case guarantees finds good clustering exists

#* cluto clutering toolkit
#@G Karypis
#t2002
#cTechnical Report
#index323
#id323
#%
#! clustering algorithms divide data meaningful useful groups called clusters intra cluster similarity maximized inter cluster similarity minimized discovered clusters used explain characteristics underlying data distribution serve foundation various data mining analysis techniques applications clustering include characterization different customer groups based purchasing patterns categorization documents world wide web grouping genes proteins similar functionality grouping spatial locations prone earth quakes seismological data cluto software package clustering low high dimensional datasets analyzing characteristics various clusters

#* chameleon hierarchical clustering using dynamic modeling
#@G Karypis,Eui-Hong Han,Vipin Kumar
#t1999
#cIEEE Computer
#index324
#id324
#%
#! clustering discovery process data mining groups set data way maximizes similarity clusters minimizes similarity different clusters advanced algorithms difficulty dealing highly variable clusters follow preconceived model basing selections interconnectivity closeness chameleon algorithm yields accurate results highly variable clusters existing algorithms use static model clusters use information nature individual clusters merged furthermore set schemes cure algorithm related schemes ignores information aggregate interconnectivity items clusters set schemes rock algorithm group averaging method related schemes ignores information closeness clusters defined similarity closest items clusters considering interconnectivity closeness algorithms select merge wrong pair clusters chameleon key feature accounts interconnectivity closeness identifying similar pair clusters chameleon finds clusters data set using phase algorithm phase chameleon uses graph partitioning algorithm cluster data items relatively small subclusters second phase uses algorithm genuine clusters repeatedly combining subclusters

#* informationtheoretic analysis hard soft assignment methods clustering
#@M Kearns,Y Mansour
#t1998
#cbook
#index325
#id325
#%
#! assignment methods heart algorithms unsupervised learning clustering particular known means expectation maximization em algorithms work study different methods assignment including hard assignments used means soft assignments used em known means minimizes distortion data em maximizes likelihood little known systematic differences behavior

#* bow toolkit statistical language modeling text retrieval classification clustering
#@A McCallum
#t1998
#cnull
#index326
#id326
#%
#! bow libbow library code useful writing statistical text analysis language modeling information retrieval programs current distribution includes library ends document classification rainbow document retrieval arrow document clustering crossbow

#* using unlabeled data improve text classification
#@Nigam,Kamal Paul
#t2001
#cnull
#index327
#id327
#%
#! key difficulty text classification learning algorithms require hand labeled examples learn accuratelyfi dissertation demonstrates supervised learning algorithms use small number labeled examples inexpensive unlabeled examples create high accuracy text classifiersfi assuming documents created parametric generative model expectation maximization em finds local maximum posteriori models classifiers

#* tutorial hidden markov models selected applications speech recognition
#@Lawrence R. Rabiner
#t2002
#cProceedings of the IEEE
#index328
#id328
#%
#! tutorial provides overview basic theory hidden markov models hmms originated baum petrie gives practical details methods implementation theory description selected applications theory distinct problems speech recognition results number original sources combined provide single source acquiring background required pursue area research author reviews theory discrete markov chains shows concept hidden states observation probabilistic function state used effectively theory illustrated simple examples coin tossing classic balls urns fundamental problems hmms noted practical techniques solving problems given various types hmms studied including ergodic left right models described

#* deterministic annealing clustering compression classification regression related optimization problems
#@K. Rose
#t1998
#cIEEE
#index329
#id329
#%
#! paperfi let place withinthe neural network perspective particularly oflearningfi area neural networks greatly benefitedfrom unique position crossroads diversescientific engineering disciplines including statisticsand probability theory physics biology control signalprocessing information theory complexity theory psychology fi neural networks provided fertilesoil infusion occasionally

#* clustering sequences hidden markov models
#@Padhraic Smyth
#t1996
#cNIPS
#index330
#id330
#%
#! paper discusses probabilistic model based approach clustering sequences using hidden markov models hmms fi problemcan framed generalization standard mixturemodel approach clustering feature spacefi primary issuesare addressedfi novel parameter initialization procedure isproposed second difficult problem determiningthe number clusters data investigatedfi experimentalresults indicate proposed

#* comparison document clustering techniques
#@Michael Steinbach,G Karypis,Vipin Kumar
#t2000
#cWorkshop on Text Mining
#index331
#id331
#%
#! paper presents results experimental study common document clustering techniques particular compare main approaches document clustering agglomerative hierarchical clustering means means used standard means algorithm variant means bisecting means hierarchicalclustering portrayed better quality clustering approach limited quadratic time complexity contrast means variants time complexity linear number documents thought produce inferior clusters means agglomerative hierarchical approaches combined best worlds results indicate bisecting means technique better standard means approach good better hierarchical approaches tested variety cluster evaluation metrics propose explanation results based analysis specifics clustering algorithms nature document data

#* cluster ensembles knowledge reuse framework combining partitions
#@Alexander Strehl,J Ghosh
#t2002
#cJournal of Machine Learning Research
#index332
#id332
#%
#! paper introduces problem combining multiple partitionings set objects single consolidated clustering accessing features algorithms determined partitionings identify application scenarios resultant knowledge reuse framework cluster ensembles cluster ensemble problem formalized combinatorial optimization problem terms shared mutual information addition direct maximization approach propose ective efficient techniques obtaining high quality combiners consensus functions combiner induces similarity measure partitionings reclusters objects second combiner based hypergraph partitioning collapses groups clusters meta clusters compete object determine combined clustering low computational costs techniques quite feasible use supra consensus function evaluates approaches objective function picks best solution given situation evaluate effectiveness cluster ensembles qualitatively different application scenarios original clusters formed based non identical sets features ii original clustering algorithms worked non identical sets objects iii common data set used main purpose combining multiple clusterings improve quality robustness solution promising results obtained situations synthetic real data sets

#* relationship based clustering visualization high dimensional data mining
#@Alexander Strehl,J Ghosh
#t2003
#cINFORMS Journal on Computing
#index333
#id333
#%
#! real life data mining paper proposes relationship based approach alleviates problems stepping curse dimensionality issue working suitable similarity space instead original high dimensional attribute spacefi intermediary similarity space suitably tailored satisfy business criteria requiring customer clusters represent comparable amounts revenuefi apply efficient scalable graph partitioning basedfififi

#* constrained clustering large database
#@
#t1000
#cEMPTY
#index334
#id334
#%
#! constrained clustering finding clusters satisfy user specified constraints highly desirable applications paper introduce constrained clustering problem traditional clustering algorithms means handle scalable constraint clustering algorithm developed study starts finding initial solution satisfies user specified constraints refines solution performing confined object movements constraints algorithm consists phases pivot movement deadlock resolution phases finding optimal solution np hard propose heuristics algorithm scale large data sets using heuristic micro cluster sharing experiments effectiveness efficiency heuristics

#* criterion functions document clustering experiments analysis
#@Ying Zhao,G Karypis
#t2001
#cTechnical Report
#index335
#id335
#%
#! recent years witnessed tremendous growth volume text documents available internet digital libraries news sources company wide intranets led increased developingmethods help users effectively navigate summarize organize information ultimategoal helping looking fast high quality document clustering algorithms play animportant role goal shown provide intuitive navigation browsing mechanismby organizing large amounts information small number meaningful clusters greatly improvethe retrieval performance cluster driven dimensionality reduction term weighting query expansion thisever increasing importance document clustering expanded range applications led developmentof number new novel algorithms different complexity quality trade offs class ofclustering algorithms relatively low computational requirements treat clustering problemas optimization process seeks maximize minimize particular clustering criterion function definedover entire clustering solution focus paper evaluate performance different criterion functions problem clusteringdocuments study involves total seven different criterion functions introduced thispaper proposed past evaluation consists comprehensive experimentalevaluation involving different datasets analysis characteristics various criterionfunctions effect clusters produce experimental results set criterionfunctions consistently outperform rest newly proposed criterion functions lead thebest overall results theoretical analysis criterion function shows relative performance dependson degree correctly operate clusters different tightness ii degree towhich lead reasonably balanced clusters

#* hmms coupled hmms multi channel eeg classification
#@Shi Zhong,J Ghosh
#t2002
#cInternational Joint Conference on Neural Networks IJCNN'02
#index336
#id336
#%581,582,328,583,584,585,586,587,588,589,590,591,592,593,594,595,596,597,330
#! variety coupled hmms chmms recently proposed extensions hmm better characterize multiple interdependent sequences paper introduces novel distance coupled hmm compares performance hmm chmm models multi channel eeg classification problem results approaches examined multivariate hmm low computational complexity surprisingly outperforms models

#* unified framework modelbased clustering applications clustering time sequences
#@Shi Zhong,J Ghosh
#t2003
#cJournal of Machine Learning Research
#index337
#id337
#%
#! model based clustering techniques widely used shown promising results applications involving complex data paper presents unified framework probabilistic model based clustering based bipartite graph view data models highlights commonalities differences existing model based clustering algorithms view clusters represented probabilistic models model space conceptually separate data space partitional clustering view conceptually similar expectationmaximization em algorithm hierarchical clustering graph based view helps visualize critical important distinctions similarity based approaches model based approaches framework suggests useful variations existing clustering algorithms new variations balanced model based clustering hybrid model based clustering discussed empirically evaluated variety data types

#* continuous time bayesian networks
#@Uri Nodelman,Christian R. Shelton,Daphne Koller
#t2002
#cUAI
#index338
#id338
#%339,340,341,342,343,344,345
#! paper present language finite state continuous time bayesian networks ctbns structured stochastic processes evolve continuous time state decomposed set local variables values change time dynamics described specifying behavior local variable function parents directed possibly cyclic graph model specifies given point time distribution aspects local variable changes value value takes distributions determined variable current value current values parents graph formally variable modelled finite state continuous time markov process transition intensities functions parents present probabilistic semantics language terms generative model ctbn defines sequences events list types queries ask ctbn discuss conceptual computational difficulties associated exact inference provide algorithm approximate inference takes advantage structure process

#* event history analysis
#@Paul David Allison
#t1984
#cbook
#index339
#id339
#%
#! purpose event history analysis explain certain individualsare higher risk experiencing event thiscan accomplished using special types methods depending onthe field applied called failure time models life timemodels survival models transition rate models response time models eventhistory models duration models hazard models examples textbooksdiscussing class techniques use terms event history survival hazard models interchange ably hazard model regression model risk experienc ing event certain time point predicted set covariates special features distinguish hazard models types regres sion models make possible deal censoredobservations contain partial information timing event special feature covariates change theirvalue observation period possibility including time varying covariates makes possible perform truly dynamic analysis discussing important types hazard models introduce basic concepts

#* model reasoning persistence causation
#@Thomas Dean,Keiji Kanazawa
#t1989
#cComputational Intelligence
#index340
#id340
#%
#! reasoning change requires predicting long proposition having true continue lacking perfect knowledge agent constrained believe proposition persists indefinitely simply way agent infer contravening proposition certainty paper model causal reasoning accounts knowledge concerning cause effect relationships knowledge concerning tendency propositions persist function time passing model natural encoding form network representation probabilistic models consider computational properties model reviewing recent advances computing consequences models encoded network representation finally discuss probabilistic model addresses certain classical problems temporal reasoning frame qualification problems

#* recursive valuation defaultable securities timing resolution uncertainty
#@Darrell Duffie,Mark Schroder,Costis Skiadas
#t1996
#cAnn. Appl. Probab.
#index341
#id341
#%
#! derive implications default risk valuation securities setting fractional default recovery rate hazard rate default depend market value instrument itself market values instruments issued entity determined simultaneously key technique use backward recursive stochastic integral equations characterize dependence market value manner resolution uncertainty particular conditions monotonicity value respect information filtration

#* probabilistic temporal reasoning endogenous change
#@Steve Hanks,David Madigan,Jonathan Gavrin
#t1995
#cUAI
#index342
#id342
#%
#! paper presents probabilistic model reasoning state changes time exogenous endogenous influencesfi target domain class medical prediction problems urgent preclude careful diagnosis progress slowly allow arbitrary testing treatment optionsfi domains typically time gather information patient state consider alternative diagnoses

#* cox processes credit risky securities
#@David Lando
#t1998
#cDerivatives Research
#index343
#id343
#%
#! framework presented modeling defaultable securities credit derivatives allows dependence market risk factors credit risk framework reduces technical issues modeling credit risk issues faced modeling ordinary term structure rates shown generalize model jarrow lando turnbull allow stochastic transition intensities rating categories default generalization handle contracts payments explicitly linked ratings shown obtain term structure model different rating categories simultaneously obtain affine like structure implementation given simple factor model affine structure gives closed form solutions

#* probabilistic reasoning intelligent systems
#@Judea Pearl
#t1988
#cbook
#index344
#id344
#%
#! probabilistic reasoning intelligent systems complete andaccessible account theoretical foundations computational methods underlie plausible reasoning uncertainty author provides coherent explication probability language reasoning partial belief offers unifying perspective ai approaches uncertainty dempster shafer formalism truth maintenance systems nonmonotonic logic author distinguishes syntactic semantic approaches uncertainty offers techniques based belief networks provide mechanism making semantics based systems operational specifically network propagation techniques serve mechanism combining theoretical coherence probability theory modern demands reasoning systems technology modular declarative inputs conceptually meaningful inferences parallel distributed computation application areas include diagnosis forecasting image interpretation multi sensor fusion decision support systems plan recognition planning speech recognition short task requiring conclusions drawn uncertain clues incomplete information probabilistic reasoning intelligent systems special scholars researchers ai decision theory statistics logic philosophy cognitive psychology management sciences professionals areas knowledge based systems operations research engineering statistics theoretical computational tools immediate practical use book used excellent text graduate level courses ai operations research applied probability

#* probability propagation
#@Glenn Shafer,Prakash P. Shenoy
#t1990
#cAnn. Math. Artif. Intell.
#index345
#id345
#%
#! paper simple account local computation marginalprobabilities joint probability distribution given factored formand sets variables involved factors form hypertree previousexpositions local computation emphasized conditional probability believe emphasis misplaced essential local computation isa factorization essential factorization interpreted terms ofconditional probabilities account given avoids divisions required byconditional probabilities generalizes readily alternative measures ofsubjective probability dempster shafer spohnian belief functions

#* correlated topic models
#@David M. Blei,J Lafferty
#t2005
#cNIPS
#index346
#id346
#%347,182,348,349,350,351,352,353,354
#! topic models latent dirichlet allocation lda useful tools statistical analysis document collections discrete data lda model assumes words document arise mixture topics distribution vocabulary limitation lda inability model topic correlation example document genetics likely disease ray astronomy limitation stems use dirichlet distribution model variability topic proportions paper develop correlated topic model ctm topic proportions exhibit correlation logistic normal distribution derive mean field variational inference algorithm approximate posterior inference model complicated fact logistic normal conjugate multinomial ctm gives better fit lda collection ocred articles journal science furthermore ctm provides natural way visualizing exploring unstructured data sets

#* vibes variational inference engine bayesian networks
#@Christopher M. Bishop,David J. Spiegelhalter,J. M. Winn
#t2002
#cNIPS
#index347
#id347
#%209,355,356,357,196,358,359
#! recent years variational methods popular tool approximate inference learning wide variety probabilistic models new application currently necessary derive variational update equations implement application specific code steps time consuming error prone paper general purpose inference engine called vibes variational inference bayesian networks allows wide variety probabilistic models implemented solved variationally recourse coding new models specified simple script graphical interface analogous drawing package vibes automatically generates solves variational equations illustrate power exibility vibes using examples bayesian mixture modelling

#* simplicial mixtures markov chains distributed modelling dynamic user profiles
#@Mark Girolami,Ata Kaban
#t2003
#cNIPS
#index348
#id348
#%360,361,362,363,364,365
#! provide compact generative representation sequential activity number individuals group tradeoff ni tion individual speci global models paper proposes linear time distributed model nite state symbolic sequences representing traces individual user activity making assumption heterogeneous user havior explained relatively small number structurally simple common behavioral patterns interleave randomly user speci proportion results empirical study different sources user traces indicates modelling approach provides ef cient representation scheme ected improved prediction performance providing low complexity intuitively interpretable representations

#* integrating topics syntax
#@Thomas L. Griffiths,Mark Steyvers,David M. Blei,Joshua B. Tenenbaum
#t2004
#cNIPS
#index349
#id349
#%366,367,368,369,182,370,371,372,373,374
#! statistical approaches language learning typically focus short range syntactic dependencies long range semantic dependencies words present generative model uses kinds dependencies capable simultaneously finding syntactic classes semantic topics despite having knowledge syntax semantics statistical dependency model competitive tasks like speech tagging document classification models exclusively use short long range dependencies respectively

#* collaborative filtering machine learning perspective
#@Benjamin Marlin,Benjamin Marlin
#t2004
#cnull
#index350
#id350
#%
#! collaborative filtering initially proposed framework filtering information based preferences users refined different ways thesis comprehensive study rating based pure non sequential collaborative filtering analyze existing methods task rating prediction machine learning perspective existing methods proposed task simple applications modifications standard machine learning methods classification regression clustering dimensionality reduction density estimation introduce new prediction methods classes introduce new experimental procedure testing stronger forms generalization used previously implement total prediction methods conduct large scale prediction accuracy experiments interesting new results relative performance methods

#* inference population structure using multilocus genotype data
#@JK Pritchard, M. Stephens, P. Donnelly 
#t1000
#cEMPTY
#index351
#id351
#%
#! model based clustering method using multilocus genotype data infer population structure assign individuals populations assume model populations unknown characterized set allele frequencies locus individuals sample assigned probabilistically populations jointly populations genotypes indicate admixed model does assume particular mutation process applied commonly used genetic markers provided closely linked applications method include demonstrating presence population structure assigning individuals populations studying hybrid zones identifying migrants admixed individuals method produce highly accurate assignments using modest numbers loci seven microsatellite loci example using genotype data endangered bird species software used article available http www stats ox ac uk zpritch home html

#* discovering object categories image collections
#@Josef Sivic,Bryan C. Russell,Alexei A. Efros,Andrew Zisserman,William T. Freeman
#t0
#cnull
#index352
#id352
#%375,376,182,377,378,379,371,205,360,380,381,382,383,384,385,386,387,388,389
#! given set images containing multiple object categories seek discover categories image locations supervision achieve using generative models statistical text literature probabilistic latent semantic analysis plsa latent dirichlet allocation lda text analysis used discover topics corpus using bag words document representation discover topics object categories image containing instances categories modelled mixture topics models applied images using visual analogue word formed vector quantizing sift like region descriptors investigate set increasingly demanding scenarios starting image sets containing object categories sets containing multiple categories including airplanes cars faces motorbikes spotted cats background clutter object categories sample intra class scale variation categories approximate spatial layout supervision demonstrate classi cation unseen images images containing multiple objects performance proposed unsupervised method compared semi supervised approach work sponsored eu project cogvisys university oxford shell oil national geospatial intelligence agency

#* variational principle graphical models
#@H Attias
#t2000
#cbook
#index353
#id353
#%
#! graphical models bring graph theory probability theory powerful formalism multivariate statistical modeling statistical signal processing related fields communication theory control theory bioinformatics statistical models long formulated terms graphs algorithms computing basic statistical quantities likelihoods marginal probabilities expressed terms recursions operating

#* generalized mean field algorithm variational inference exponential families
#@Eric P. Xing,Michael I. Jordan,Stuart J. Russell
#t2003
#cUAI
#index354
#id354
#%
#! present class generalized mean field gmf algorithms approximate inference exponential family graphical models analogous generalized belief propagation gbp cluster variational methodsfi methods based

#* new view em algorithm justifies incremental variants
#@Radford Neal,Geoffrey E. Hinton
#t1993
#cLearning in Graphical Models
#index355
#id355
#%
#! present new view em algorithm maximum likelihood estimationin situations unobserved variablesfi view stepsof algorithm seen maximizing joint function model parametersand distribution unobserved variablesfi perspective iseasy justify incremental variant algorithm distributionfor unobserved variables recalculated stepfi thisvariant shown

#* winbugs bayesian modelling framework concepts structure extensibility
#@David J. Lunn,Andrew Thomas,Nicky Best,David Spiegelhalter
#t2000
#cStatistics and Computing
#index356
#id356
#%
#! winbugs fully extensible modular framework constructing analysing bayesian probability models models specified textually bugs language pictorially using graphical interface called doodlebugs winbugs processes model specification constructs object oriented representation model software offers user interface based dialogue boxes menu commands model analysed using markov chain monte carlo techniques paper discuss various modern computing concepts object orientation run time linking feature software design discuss framework extended possible write specific applications form apparently seamless interface winbugs users specialized requirements possible interface winbugs lower level incorporating new object types used winbugs knowledge modules implemented types extension require access recompilation winbugs source code

#* propagation algorithms variational bayesian learning
#@Zoubin Ghahramani,Matthew J. Beal
#t2000
#cNIPS
#index357
#id357
#%
#! variational approximations widespread tool forbayesian learning graphical modelsfi provide theoreticalresults variational updates general family ofconjugate exponential graphical modelsfi beliefpropagation junction tree algorithms used inference step variational bayesian learningfi applying resultsto bayesian analysis linear gaussian state space modelswe obtain learning procedure

#* variational principal components
#@Christopher M. Bishop
#t1999
#cProceedings Ninth International Conference on Artificial Neural Networks, ICANN99
#index358
#id358
#%
#! central issues use princi pal component analysis pca data mod elling choosing appropriate num ber retained components problem wasrecently addressed formulation abayesian treatment pca bishop interms probabilistic latent variable model acentral feature approach effec tive dimensionality latent space equiv alent number retained principal com ponents determined automatically partof bayesian inference procedure com mon non trivial bayesian models required marginalizations alytically intractable approximationscheme based local gaussian representa tion posterior distribution employed paper develop alternative varia tional formulation bayesian pca based ona factorial representation posterior distri bution approach computationally effi cient unlike approximation schemes maximizes rigorous lower bound themarginal log probability observed data

#* structured variational distributions vibes
#@Christopher M. Bishop,J. M. Winn
#t2003
#cProceedings Artificial Intelligence and Statistics
#index359
#id359
#%
#! variational methods increasinglypopular approximate solution complexprobabilistic models machine learning vision information retrieval manyother fieldsfi unfortunately new applicationit necessary derive specificforms variational update equations forthe particular probabilistic model used andthen implement equations applicationspecificsoftwarefi steps bothtime

#* unsupervised learning probabilistic latent semantic analysis
#@Thomas Hofmann
#t2001
#cMachine Learning
#index360
#id360
#%
#! paper presents novel statistical method factor analysis binary count data whichis closely related technique known latent semantic analysis contrast method whichstems linear algebra performs singular value decomposition occurrence tables proposedtechnique uses generative latent class model perform probabilistic mixture decomposition resultsin principled approach solid foundation statistical inference precisely propose tomake use temperature controlled version expectation maximization algorithm model fitting whichhas shown excellent performance practice probabilistic latent semantic analysis applications mostprominently information retrieval natural language processing machine learning text related areas paper presents perplexity results different types text linguistic data collections discusses anapplication automated document indexing experiments indicate substantial consistent improvementsof probabilistic method standard latent semantic analysis

#* maximum likelihood estimation dirichlet distributions
#@G Ronning
#t1989
#cJournal of statistical computation and simulation
#index361
#id361
#%682,182
#! dirichlet distributions commonly used priors proportional data paper introduce distribution discuss useful compare implementations different methods estimating parameters observed data

#* maximum likelihood estimation mover stayer model
#@Halina Frydman
#t1984
#cJournal of the American Statistical Association
#index362
#id362
#%
#! discrete time mover stayer model special mixture independent markov chains widely used modeling dynamics social processes problem maximum likelihood estimation parameters data consist sample independent realizations process considered literature present maximum likelihood procedure estimation parameters mover stayer model develop recursive method computation maximum likelihood estimators simple implement verify obtained maximum likelihood estimators strongly consistent estimators parameters mover stayer model previously proposed literature special cases maximum likelihood estimator derived article coincide maximum likelihood estimator special conditions explain interconnection existing estimators present numerical comparison estimators finally illustrate application maximum likelihood estimators testing hypothesis markov chain describes data hypothesis mover stayer model describes data

#* algorithms non negative matrix factorization
#@Daniel D. Lee,H. Sebastian Seung
#t2000
#cNIPS
#index363
#id363
#%
#! non negative matrix factorization nmf previously shown useful decomposition multivariate data different multi plicative algorithms nmf analyzed differ slightly multiplicative factor used update rules algorithm shown minimize conventional squares error minimizes generalized kullback leibler divergence monotonic convergence algorithms proven using auxiliary func tion analogous used proving convergence expectation maximization algorithm algorithms interpreted diag onally rescaled gradient descent rescaling factor optimally chosen ensure convergence

#* expectation propogation generative aspect model
#@Thomas P. Minka,J Lafferty
#t2002
#cUAI
#index364
#id364
#%
#! generative aspect model extension ofthe multinomial model text allows wordprobabilities vary stochastically documentsfi

#* ensemble learning
#@Robi Polikar
#t2009
#cnull
#index365
#id365
#%
#! introductionwhen say making model setting tool canbe used make inferences predictions decisionsfi model seen ahypothesis explanation makes assertions quantities aredirectly observable inferred effect observablequantitiesfiin bayesian framework knowledge contained conditional probabilitydistributions modelsfi use bayes theorem

#* fractionating language different neural subsytems different sensitive periods
#@HJ Neville,DL Mills,DS Lawson
#t1992
#cCerebral Cortex
#index366
#id366
#%
#! theoretical considerations psycholinguistic studies alternately provided criticism support proposal semantic grammatical functions distinct subprocesses language domain neurobiological evidence concerning hypothesis sought comparing normal adults event related brain potentials erps elicited words provide primarily semantic information open class grammatical information closed class comparing effects altered early language experience congenitally deaf subjects erps open closed class words normal hearing adults different word types elicited qualitatively different erps compatible hypothesized different roles word classes language processing addition erp indices semantic processing virtually identical deaf hearing subjects linked grammatical processes markedly different deaf bearing subjects results suggest nonidentical neural systems different developmental vulnerabilities mediate different aspects language generally results provide neurobiological support distinction semantic grammatical functions

#* distributional information powerful cue acquiring syntactic categories
#@Martin Redington,Nick Chater,Steven Finch
#t1998
#cCognitive Science
#index367
#id367
#%
#! theorists dismissed priori idea distributional information play significant role syntactic category acquisition demonstrate empirically information provides powerful cue syntactic category membership exploited variety simple psychologically plausible mechanisms present range results using large corpus child directed speech explore psychological implications results considerable information concerning syntactic categories obtained distributional information stress sources information potential contributors identification syntactic classes

#* solution plato problem latent semantic analysis theory acquisition induction representation knowledge
#@TK Landauer,S. Dumais
#t1997
#cPsychological Review
#index368
#id368
#%
#! people know little information problem takes forms learning vocabulary text especially dramatic convenient case research new general theory acquired similarity knowledge representation latent semantic analysis lsa presented used successfully simulate learning psycholinguistic phenomena inducing global knowledge indirectly local occurrence data large body representative text lsa acquired knowledge vocabulary english comparable rate school children lsa uses prior linguistic perceptual similarity knowledge based solely general mathematical learning method achieves powerful inductive effects extracting right number dimensions represent objects contexts relations theories phenomena problems sketched

#* foundations statistical natural language processing
#@Paul B. Kantor
#t2001
#cInf. Retr.
#index369
#id369
#%
#! statistical approaches processing natural language text dominant recent years foundational text comprehensive introduction statistical natural language processing nlp appear book contains theory algorithms needed building nlp tools provides broad rigorous coverage mathematical linguistic foundations detailed discussion statistical methods allowing students researchers construct implementations book covers collocation finding word sense disambiguation probabilistic parsing information retrieval applications

#* better integration semantic predictors statistical language modeling
#@Noah Coccaro,Daniel Jurafsky
#t0
#c1998
#index370
#id370
#%620,66,621,622,623,624,625
#! introduce number techniques designed help integrate semantic knowledge gram language models automatic speech recognition techniques allow integrate latent semantic analysis lsa word similarity algorithm based word occurrence information gram models lsa good predicting content words coherent rest text bad predictor frequent words low dynamic range inaccurate combined linearly grams modifying dynamic range applying word confidence metric using geometric linear combinations grams produces robust language model lower perplexity wall street journal testset baseline gram model

#* finding scientific topics
#@Thomas L. Griffiths, Mark Steyvers
#t2004
#cProc Natl Acad Sci USA
#index371
#id371
#%
#! step identifying content document determining topics document addresses generative model documents introduced blei ng jordan document generated choosing distribution topics choosing word document topic selected according distribution present markov chain monte carlo algorithm inference model use algorithm analyze pnas using bayesian model selection establish number topics extracted topics capture meaningful structure data consistent class designations provided authors articles outline applications analysis including identifying hot topics examining temporal dynamics tagging abstracts illustrate semantic content

#* markov chain monte carlo practice
#@WR Gilks,DJ Spiegelhalter
#t1996
#c1996	book
#index372
#id372
#%
#! markov chain monte carlo mcmc methods make possi ble use flexible bayesian models wise computationally infeasible recent years greatvariety applications described lit erature applied statisticians new methodsmay questions concerns howmuch effort expertise needed design use amarkov chain sampler confidence havein answers mcmc produces does use ofmcmc affect rest model building process atthe joint statistical meetings august panel ofexperienced mcmc users discussed issues various tricks trade article anedited recreation discussion purpose offeradvice guidance novice users mcmcand novice users topics include building confidencein simulation results methods speeding assessingconvergence estimating standard errors identification ofmodels good mcmc algorithms exist thecurrent state software development

#* bayes factors
#@Robert E. Kass, Adrian E. Raftery 
#t1995
#cJournal of the American Statistical Association
#index373
#id373
#%
#! paper book theory probability jeffreys developed methodologyfor quantifying evidence favor scientific theory centerpiece number called bayes factor posterior odds null hypothesis priorprobability null halffi discussion bayesianhypothesis testing context criticism values attention given thebayes factor practical tool

#* comparing partitions
#@Lawrence Hubert,Phipps Arabie
#t1985
#cJournal of Classification
#index374
#id374
#%
#! problem comparing different partitions finite set objects reappears continually clustering literature begin reviewing known measure partition correspondence attributed rand discuss issue correcting index chance note recent normalization strategy developed morey agresti adopted miligan cooper based incorrect assumption general problem comparing partitions approached indirectly assessing congruence proximity matrices using simple cross product measure generated corresponding partitions using various scoring rules special cases derivable include traditionally familiar statistics ones tailored weight certain object pairs differentially finally propose measure based comparison object triples having advantage probabilistic interpretation addition corrected chance assuming constant value reasonable null hypothesis bounded

#* matching words pictures
#@Kobus Barnard,Pinar Duygulu,David A. Forsyth,Nando de Freitas,David M. Blei,Michael I. Jordan
#t2003
#cJournal of Machine Learning Research
#index375
#id375
#%
#! present new rich approach moeling multi moal ata sets focusing specific case segmente images associate textfi learning joint istribution image regions wors applicationsfi consier inetail preicting wors associate images auto annotation ancorresponing particular image regions region naming fi auto annotation mighthelp organize access large collections imagesfi region naming moel ofobject

#* minimum complexity density estimation
#@Andrew R. Barron,TM Cover
#t1991
#cIEEE Transactions on Information Theory
#index376
#id376
#%
#! authors introduce index resolvability proved bound rate convergence minimum complexity density estimators information theoretic redundancy corresponding total description length results index resolvability demonstrate statistical effectiveness minimum description length principle method inference minimum complexity estimator converges true density nearly fast estimator based prior knowledge true subclass densities interpretations basic properties minimum complexity estimators discussed regression classification problems examined minimum description length framework considered

#* visual catego rization bags keypoints
#@Gabriella Csurka,Christopher R. Dance,Lixin Fan,Jutta Willamowski,Cdric Bray
#t2004
#cWorkshop on Statistical Learning in Computer Vision, ECCV (2004)
#index377
#id377
#%
#! present novel method generic visual categorization problem identifying object content natural images generalizing variations inherent object class bag keypoints method based vector quantization affine invariant descriptors image patches propose compare alternative implementations using different classifiers na ve bayes svm main advantages method simple computationally efficient intrinsically invariant present results simultaneously classifying seven semantic visual categories results clearly demonstrate method robust background clutter produces good categorization accuracy exploiting geometric information

#* object recognition machine translation learning lexicon fixed image vocabulary
#@Pinar Duygulu,Kobus Barnard,Joao F. G. de Freitas,David A. Forsyth
#t2002
#cECCV (4)
#index378
#id378
#%
#! model object recognition machine translationfi

#* learning generative visual models training examples incremental bayesian approach tested object categories
#@Fei-Fei Li,Robert Fergus,Pietro Perona
#t2007
#cComputer Vision and Image Understanding
#index379
#id379
#%
#! current computational approaches learning visual object categories require thousands training images slow learn incremental manner incorporate prior information learning process addition algorithm presented literature tested handful object categories present method learning object categories just training images quick uses prior information principled way test dataset composed images objects belonging widely varied categories proposed method based making use prior information assembled unrelated object categories previously learnt generative probabilistic model used represents shape appearance constellation features belonging object parameters model learnt incrementally bayesian manner incremental algorithm compared experimentally earlier batch bayesian algorithm based maximum likelihood incremental batch versions comparable classification performance small training sets incremental learning significantly faster making real time learning feasible bayesian methods outperform maximum likelihood small training sets

#* robust wide baseline stereo maximally stable extremal regions
#@Jiri Matas,Ondrej Chum,Martin Urban,Tomas Pajdla
#t2002
#cBMVC
#index380
#id380
#%
#! wide baseline stereo problem problem establishing correspon dences pair images taken different viewpoints studied new set image elements correspondence called extremal regions introduced sirable properties set closed transformation image coordinates age intensities efficient near linear complexity practically fast tection algorithm near frame rate presented affinely subset extremal regions maximally stable extremal regions mser new robust similarity measure establishing tentative correspon dences proposed robustness ensures invariants multiple measurement regions regions obtained invariant constructions ex tremal regions significantly larger discriminative msers used establish tentative correspondences high utility msers multiple measurement regions robust metric demonstrated wide baseline experiments image pairs indoor outdoor scenes significant change scale nation conditions plane rotation occlusion locally anisotropic scale change translation viewpoint present test prob lems good estimates epipolar geometry average distance corre sponding points epipolar line obtained

#* weak hypotheses boosting generic object detection recognition
#@Andreas Opelt,Michael Fussenegger,Axel Pinz,Peter Auer
#t2004
#cECCV (2)
#index381
#id381
#%
#! fi paper stage new learning object detection recognitionfi propose boosting underlying learning techniquefi allows use diverse sets visual features learning process com mon framework boosting weak hypotheses finder choose inhomogeneous features relevant combina tion final hypothesisfi advantage weak hypotheses finder search weak hypotheses space explicit calculation available hypotheses reducing computation timefi contrasts related work agarwal roth winnow used learning algorithm weak hypotheses calculated explicitlyfi empirical evaluation use types local descriptors basic ones consisting set grayvalues intensity moments high level descriptors moment invariants sifts fi descriptors calculated local patches detected inter est point operatorfi weak hypotheses finder selects local patches type local descriptor ciently searches discriminative similarity thresholdfi di ers work boosting object recognition simple rectangular hypotheses complex classifiers usedfi relatively simple images objects prominent approach yields results comparable state art fi obtain good results complex images objects located arbitrary positions poses scales imagesfi results indicate flexible approach allows inclusion features segmented gions spatial relationships leads significant step generic object recognitionfi

#* statistical method object detection applied faces cars
#@Henry Schneiderman,Takeo Kanade
#t2000
#cCVPR
#index382
#id382
#%
#! paper statistical method object detectionfi represent statistics object appear ance non object appearance using product histo gramsfi histogram represents joint statistics subset wavelet coefficients position objectfi approach use histograms representing wide variety visual attributesfi using method devel oped algorithm reliably detect human faces plane rotation algorithm reli ably detect passenger cars wide range viewpointsfi

#* video google text retrieval approach object matching videos
#@Josef Sivic,Andrew Zisserman
#t2003
#cICCV
#index383
#id383
#%
#! approach object scene retrieval searches localizes occurrences user outlined object video object represented set viewpoint invariant region descriptors recognition proceed successfully despite changes viewpoint illumination partial occlusion temporal continuity video shot used track regions order reject unstable regions reduce effects noise descriptors analogy text retrieval implementation matches descriptors pre computed using vector quantization inverted file systems document rankings used result retrieval immediate returning ranked list key frames shots manner google method illustrated matching length feature films

#* automated image retrieval using color texture
#@J.R. Smith,S.F. Chang
#t1995
#cTech. Rep.
#index384
#id384
#%
#! growing prevalence digital images videos increasing need filters based visual content unfortunately appearance effective tools searching retrieving images videos large archives accompanied proliferation images videos digital form addition text based indexes built human supplied annotations digital image video libraries require new algorithms automated extraction indexing relevant image features paper investigate automated content extraction database searching based color texture features features important colors textures fundamental characteristics content images giving work general application databases images videos variety domains present new algorithms automated extraction color texture information use binary set representations color texture respectively special emphasis processes feature extraction indexing demonstrate binary feature sets texture color provide excellent performance query response time maintaining highly effective discriminability accuracy spatial localization capability extraction compressed data representations present color texture set extraction indexing techniques contrast approaches examine texture color searching databases color images respectively finally examine relationship color texture application image video retrieval explore capability combining color texture maps images obtain unified feature maps characterize image regions color texture particular indicate combining modalities better capture index color patterns characterize real world objects images videos finally examine nature performance image video database queries combine color texture

#* improved video content indexing multiple latent semantic analysis
#@Fabrice Souvannavong,Bernard Merialdo,Benoit Huet
#t2004
#cCIVR
#index385
#id385
#%
#! low level features insufficient build efficient content based retrieval systems users interested longer retrieving visually similar content expect retrieval systems documents similar semantic content bridging gap low level features semantic content challenging task necessary future retrieval systems latent semantic analysis lsa successfully introduced efficiently index text documents detecting synonyms polysemy words successfully proposed adaptation lsa model video content object retrieval semantic content estimation following idea present new model composed multiple lsas lsa better represent video content experimental section make comparison lsa lsa problems object retrieval semantic content estimation

#* contextual priming object detection
#@Antonio B. Torralba
#t2003
#cInternational Journal of Computer Vision
#index386
#id386
#%
#! general consensus context rich source information object identity location scalefi fact structure real world scenes governed strong configurationalrules akin apply single objectfi introduce simple framework modelingthe relationship context object properties based correlation statisticsof low level features entire scene objects containsfi

#* rapid object detection using boosted cascade simple features
#@Paul A. Viola,Michael J. Jones
#t2001
#cCVPR (1)
#index387
#id387
#%
#! paper describes machine learning approach visual object detection capable processing images extremely rapidly achieving high detection ratesfi work distinguished key contributionsfi introduction new image representation called integral image allows features used detector computed quicklyfi second learning algorithm based adaboost selects small number critical visual features

#* unsupervised learning models recognition
#@Markus Weber,Max Welling,Pietro Perona
#t2000
#cECCV (1)
#index388
#id388
#%
#! fi present method learn object class models unlabeled andunsegmented cluttered scenes purpose visual object recognitionfi wefocus particular type model objects represented flexible constellationsof rigid parts features fi variability class representedby joint probability density function pdf shape constellation andthe output detectorsfi stage method automatically identifiesdistinctive

#* hidden semantic concept discovery region based image retrieval
#@Ruofei Zhang,Zhongfei (Mark) Zhang
#t2004
#cCVPR (2)
#index389
#id389
#%
#! work addresses content based image retrieval cbir focusing developing hidden semantic concept discovery methodology address effective semantics intensive image retrieval approach image database segmented region associated homogenous color texture shape features exploiting regional statistical information image employing vector quantization method uniform sparse region based representation achieved representation probabilistic model based statistical hidden class assumptions image database obtained expectation maximization em technique applied analyze semantic concepts hidden database elaborated retrieval algorithm designed support probabilistic model semantic similarity measured integrating posterior probabilities transformed query image constructed negative example discovered semantic concepts proposed approach solid statistical foundation experimental evaluations database general purposed images demonstrate promise effectiveness

#* designing novel review ranking systems
#@Anindya Ghose,Panagiotis G. Ipeirotis
#t2007
#cACM International Conference Proceeding Series
#index390
#id390
#%391,392,393,394,395,185,396,143,397,146,195,398,399
#! rapid growth internet users ability publish content created active electronic communities provide wealth product information consumers naturally gravitate reading reviews order decide buy product high volume reviews typically published single product makes harder individuals locate best reviews understand true underlying quality product based reviews similarly manufacturer product needs identify reviews inuence customer base examine content reviews paper propose ranking mechanisms ranking product reviews consumer oriented ranking mechanism ranks reviews according expected helpfulness manufactureroriented ranking mechanism ranks reviews according expected ect sales ranking mechanism combines econometric analysis text mining techniques general subjectivity analysis particular subjectivity analysis useful clues helpfulness review impact sales results implications market design online opinion forums

#* money deriving pricing power product features mining consumer reviews
#@Nikolay Archak,Anindya Ghose, Panagiotis Ipeirotis
#t2007
#cKDD
#index391
#id391
#%
#! increasing pervasiveness internet dramatically changed way consumers shop goods consumergenerated product reviews valuable source information customers read reviews decide buy product based information provided paper use techniques decompose reviews segments evaluate individual characteristics product image quality battery life digital camera major contribution paper adapt methods econometrics literature specifically hedonic regression concept estimate weight customers place individual product feature implicit evaluation score customers assign feature evaluations ect revenue given product goal develop novel hybrid technique combining text mining econometrics models consumer product reviews elements tensor product feature evaluation spaces impute quantitative impact consumer reviews product demand linear functional tensor product space demonstrate use lowdimension approximation functional signicantly reduce number model parameters providing good experimental results evaluate technique using data set amazon com consisting sales data related consumer reviews posted month period products experimental evaluation shows extract actionable business intelligence data better understand customer preferences actions textual portion reviews improve product sales prediction compared baseline technique simply relies numeric data

#* effect word mouth sales online book reviews
#@JA Chevalier,D Mayzlin
#t2006
#cJournal of Marketing Research
#index392
#id392
#%
#! creation online consumer communities provide product reviews advice touted important albeit somewhat expensive component internet retail strategies paper characterize reviewer behavior popular internet sites examine effect consumer reviews firms sales use publicly available data leading online booksellers amazon com barnesandnoble com construct measures firm sales individual books gather extensive consumer review data sites characterize reviewer behavior sites distribution number ratings valence length ratings ratings different subject categories second measure effect individual reviews relative shares books sites argue methodology comparing sales reviews given book internet retailers allows improve existing literature better capturing causal relationship word mouth reviews sales able difference factors affect sales word mouth retailers book quality examine incremental sales effects having reviews particular book versus having reviews differential sales effects positive negative reviews large database books allows control important confounding factors differences sites prices shipping times

#* yahoo amazon sentiment extraction small talk web
#@SR Das,MY Chen
#t2007
#cmangement science
#index393
#id393
#%
#! extracting sentiment text hard semantic problem develop methodology extracting small investor sentiment stock message boards algorithm comprises different classifier algorithms coupled voting scheme accuracy levels similar widely used bayes classifiers false positives lower sentiment accuracy higher time series cross sectional aggregation message information improves quality resultant sentiment index particularly presence slang ambiguity empirical applications evidence relationship stock valuestech sector postings related stock index levels volumes volatility algorithms used assess impact investor opinion management announcements press releases party news regulatory changes

#* multi level examination impact social identities economic transactions electronic markets
#@Forman Chris,Ghose Anindya,Wiesenfeld Batia
#t2006
#cSSRN
#index394
#id394
#%
#! important uses internet today economic marketplace forum social interaction source information paper explore activities come form emergent social communities built information exchanges enabled electronic marketplaces drawing social identity theory suggest relationship online consumer reviews internet product sales partially explained social identity processes using unique dataset based chronologically compiled ratings reviewer characteristics given set products geographical location based purchasing behavior amazon provide evidence community level linking prevalence identity claiming behavior online community subsequent product sales addition reviewers claim particular geographic location subsequent product sales higher region review level analysis subsequent reviews conform identity claiming norms set previous reviews identity claiming conforms community norms elicits identity granting furthermore results suggest prevalence identity granting implications economic exchange form product sales implications research word mouth electronic communities discussed

#* opinion mining using econometrics case study reputation systems
#@Anindya Ghose,Panagiotis G. Ipeirotis,Arun Sundararajan
#t2007
#cACL
#index395
#id395
#%
#! deriving polarity strength opinions important research topic attracting significant attention years work measure strength polarity opinion consider economic context opinion evaluated instead using human annotators linguistic resources rely fact text line systems influences behavior humans effect observed using easy measure economic variables revenues product prices reversing logic infer semantic orientation strength opinion tracing changes associated economic variable effect use econometrics identify economic value text assign dollar value opinion phrase measuring sentiment effectively need manual labeling argue interpreting opinions using econometrics objective quantifiable contextsensitive evaluation opinions make discussion concrete presenting results reputation amazon com user feedback affects pricing power merchants measuring pricing power infer polarity strength underlying feedback postings

#* mining summarizing customer reviews
#@Minqing Hu,Bing Liu
#t2004
#cKDD
#index396
#id396
#%400,401,402,403,404,405,406,407,408,409,410,411,412,413,414,415,416,369,417,149,418,419,420
#! merchants selling products web ask customers review products purchased associated services commerce popular number customer reviews product receives grows rapidly popular product number reviews hundreds thousands makes difficult potential customer read make informed decision purchase product makes difficult manufacturer product track manage customer opinions manufacturer additional difficulties merchant sites sell product manufacturer normally produces kinds products research aim summarize customer reviews product summarization task different traditional text summarization features product customers expressed opinions opinions positive negative summarize reviews selecting subset rewrite original sentences reviews capture main points classic text summarization task performed steps mining product features commented customers identifying opinion sentences review deciding opinion sentence positive negative summarizing results paper proposes novel techniques perform tasks experimental results using reviews number products sold online demonstrate effectiveness techniques

#* market distortions agents better informed value information real estate transactions
#@Steven D. Levitt,Chad Syverson
#t2005
#cNBER Working Paper
#index397
#id397
#%
#! agents better informed clients hire exploit informational advantage real estate agents know housing market typical homeowner example real estate agents receive small share incremental profit house sells higher value incentive convince clients sell houses cheaply quickly test predictions comparing home sales real estate agents hired sell home instances real estate agent sells home case agent distorted incentives case agent wants pursue best consistent theory homes owned real estate agents sell percent houses stay market days longer controlling wide range housing characteristics situations agent informational advantage larger lead greater distortions

#* sentiment classication using machine learning techniques
#@Bo Pang,Lillian Lee,Shivakumar Vaithyanathan
#t2002
#cCoRR
#index398
#id398
#%
#! consider problem classifying documents topic overall sentiment determining review positive negative using movie reviews data standard machine learning techniques definitively outperform human produced baselines machine learning methods employed naive bayes maximum entropy classification support vector machines perform sentiment classification traditional topic based categorization conclude examining factors make sentiment classification problem challenging

#* thumbs thumbs semantic orientation applied unsupervised classication reviews
#@Peter Turney
#t2002
#cProceedings of the 40th Annual Meeting of the Association for Computational Linguistics
#index399
#id399
#%
#! paper presents simple unsupervised learning algorithm classifying reviews recommended thumbs recommended thumbs classification review predicted average semantic orientation phrases review contain adjectives adverbs phrase positive semantic orientation good associations subtle nuances negative semantic orientation bad associations cavalier paper semantic orientation phrase calculated mutual information given phrase word excellent minus mutual information given phrase word poor review classified recommended average semantic orientation phrases positive algorithm achieves average accuracy evaluated reviews epinions sampled different domains reviews automobiles banks movies travel destinations accuracy ranges automobile reviews movie reviews

#* recognizing subjectivity case study manual tagging
#@Rebecca Bruce,Janyce Wiebe
#t1999
#cNatural Language Engineering
#index400
#id400
#%
#! paper case study sentence level categorization tagging instructions developed used judges classify clauses wall street journal subjective objective agreement judges analyzed based analysis clause given final classification provide empirical support classifications correlations assessed data subjective category basic semantic class

#* combining low level summary representations opinions multi perspective question answering
#@Claire Cardie,Janyce Wiebe,Theresa Wilson,Diane J. Litman
#t2003
#cNew Directions in Question Answering
#index401
#id401
#%
#! recent progress research fact based question answering work aims extend question answering research different direction handle multi perspective question answering tasks question answering tasks require ability organize opinions text particular paper proposes approach multi perspective question answering views task opinion oriented information extraction

#* word association norms mutual information lexicography
#@Kenneth Ward Church,Patrick Hanks
#t1989
#cACL
#index402
#id402
#%
#! term word association used particular sense psycholinguistic literature generally speaking subjects respond quicker normal word nurse follows highly associated word doctor extend term provide basis statistical description variety interesting linguistic phenomena ranging semantic relations doctor nurse type content word content word lexico syntactic occurrence constraints verbs prepositions content word function word paper propose objective measure based information theoretic notion mutual information estimating word association norms readable corpora standard method obtaining word association norms testing thousand subjects words costly unreliable proposed measure association ratio estimates word association norms directly readable corpora making possible estimate norms tens thousands words

#* study implementation combined techniques automatic extraction terminology
#@Batrice Daille
#t1996
#cThe Balancing Act: Combining Symbolic and Statistical Approaches to Language
#index403
#id403
#%
#! paper presents original method implementation extract terminology corpora combining linguistic filters statistical methods starting linguistic study terms telecommunication domain designed number filters enable obtain selection sequences considered terms various statistical scores applied selection results evaluated method applied french english

#* mining peanut gallery opinion extraction semantic classification product reviews
#@K Dave,S Lawrence,DM Pennock
#t2003
#cWWW
#index404
#id404
#%
#! web contains wealth product reviews sifting daunting task ideally opinion mining tool process set search results given item generating list product attributes quality features aggregating opinions poor mixed good begin identifying unique properties problem develop method automatically distinguishing positive negative reviews classifier draws information retrieval techniques feature extraction scoring results various metrics heuristics vary depending testing situation best methods work better traditional machine learning operating individual sentences collected web searches performance limited noise ambiguity context complete web based tool aided simple method grouping sentences attributes results qualitatively quite useful

#* wordnet electronic lexical database
#@Christiane Fellbaum
#t1998
#cbook
#index405
#id405
#%
#! wordnet important widely used lexical resource natural language processing systems wordnet electronic lexical database edited christiane fellbaum discusses design wordnet theoretical historical perspectives provides date description lexical database presents set applications wordnet book contains foreword george miller introduction christiane fellbaum seven chapters cognitive sciences laboratory princeton university wordnet produced chapters contributed scientists

#* genre classification domain transfer information filtering
#@Aidan Finn,Nicholas Kushmerick,Barry Smyth
#t2002
#cECIR
#index406
#id406
#%
#! world wide web vast repository information sheer volume makes dificult identify useful documents identify document genre important factor retrieving useful documents focus novel document genre dimension subjectivity investigate approaches automatically classifying documents genre traditional bag words techniques speech statistics hand crafted shallow linguistic features particularly interested domain transfer learned classifiers generalize training corpus new document corpus experiments demonstrate speech approach better traditional bag words techniques particularly domain transfer conditions

#* summarizing text documents sentence selection evaluation metrics
#@Jade Goldstein,Mark Kantrowitz,Vibhu O. Mittal,Jaime G. Carbonell
#t1999
#cSIGIR
#index407
#id407
#%
#! human quality text summarization systems difficult design difficult evaluate documents differ dimensions length writing style lexical usage certain cues help suggest selection sentences inclusion summary paper presents analysis news article summaries generated sentence selection sentences ranked potential inclusion summary using weighted

#* effects adjective orientation gradability sentence subjectivity
#@Vasileios Hatzivassiloglou,Janyce Wiebe
#t2000
#cCOLING
#index408
#id408
#%
#! subjectivity pragmatic sentence level feature important implications texl processing applicalions information exlractiou information iclricwd study tile elfeels dymunic adjectives semantically oriented adjectives gradable ad ieclivcs simple subjectivity classiiicr establish lhat lhcy arc strong predictors subjectivity novel trainable mclhod thai statistically combines indicators gradability presented ewlhlalcd complementing exisling automatic icchniques assigning orientation labels

#* direction based text interpretation information access refinement
#@Marti A. Hearst
#t1992
#cText-Based Intelligent Systems: Current Research and Practice in Information Extraction and Retrieval
#index409
#id409
#%
#! text based intelligent provide depth information contents corpus does standard information retrieval time avoiding complexity resource consuming behavior detailed text understanders instead focusing discovering documents pertain topic user approach introduced based criterion directionality agent favor neutral opposed event method described coercing sentence meanings metaphoric model semantic interpretation needed order determine directionality sentence respect model interpretation method designed integrated component hybrid information access

#* mining opinion features customer reviews
#@Minqing Hu,Bing Liu
#t2004
#cAAAI
#index410
#id410
#%
#! common practice merchants selling products web ask customers review products associated services commerce popular number customer reviews product receives grows rapidly popular product number reviews hundreds makes difficult potential customer read order make decision buy product project aim summarize customer reviews product summarization task different traditional text summarization interested specific features product customers opinions opinions positive negative summarize reviews selecting rewriting subset original sentences reviews capture main points classic text summarization paper focus mining opinion product features reviewers commented number techniques presented features experimental results techniques highly effective

#* fuzzy typing document management
#@Alison Huettner,Pero Subasic
#t1000
#cEMPTY
#index411
#id411
#%
#! prototype demonstrates novel method document analysis management based combination techniques nlp fuzzy logic central technique use nlp semantic typing refer approach fuzzy typing document management

#* term extraction automatic indexing
#@Ruslan Mitkov
#t2000
#cbook
#index412
#id412
#%
#! chapter presents new domain research development natural language processing nlp concerned representation acquisition recognition terms terms pervasive scientific technical documents identification crucial issue application dealing analysis understanding generation translation documents particular growing mass specialized documentation available line industrial

#* recognizing text genres simple metrics using discriminant analysis
#@Jussi Karlgren,Douglas R. Cutting
#t1994
#cCOLING
#index413
#id413
#%
#! simple method categorizing texts pre determined text genre categories using statistical standard technique discriminant analysis demonstrated application brown corpus discriminant analysis makes possible use large number parameters specific certain corpus information stream combine small number functions parameters weighted basis useful discriminating text genres application information retrieval discussed

#* automatic detection text genre
#@Brett Kessler,Geoffrey Nunberg,Hinrich Schutze
#t1997
#cACL
#index414
#id414
#%
#! text databases available users larger heterogeneous genre increasingly important computational linguistics complement topical structural principles classification propose theory genres bundles facets correlate various surface cues argue genre detection based surface cues successful detection based deeper structural properties

#* integrating classification association rule mining
#@Bing Liu,Wynne Hsu,Yiming Ma
#t1998
#cKDD
#index415
#id415
#%
#! concept lattice efficient tool data analysis paper classification association rule mining unified concept lattice framework present fast algorithm extract association classification rules concept lattice

#* multi document summarization graph search matching
#@Inderjeet Mani,Eric Bloedorn
#t1997
#cAAAI/IAAI
#index416
#id416
#%
#! new method summarizing similarities differences pair related documents using graph representation text concepts denoted words phrases proper names document represented positionally nodes graph edges corresponding semantic relations items given perspective terms pair documents summarized algorithm uses spreading activation technique discover document nodes semantically related topic activated graphs document matched yield graph corresponding similarities differences pair rendered natural language evaluation techniques carried

#* introduction wordnet line lexical database
#@George A. Miller,Richard Beckwith,Christiane Fellbaum,Derek Gross,Katherine J. Miller
#t1990
#cInt J Lexicography
#index417
#id417
#%
#! wordnet line lexical reference design inspired current psycholinguistic theories human lexical memory english nouns verbs adjectives organized synonym sets representing underlying lexical concept different relations link synonym sets

#* automatic text decomposition using text segments text themes
#@Gerard Salton,Amit Singhal,Chris Buckley,Mandar Mitra
#t1996
#cHypertext
#index418
#id418
#%
#! widespread use text information retrieval passage retrieval techniques increasingly popular larger texts replaced important text excerpts simplifying retrieval task improving retrieval effectiveness passagelevel evidence use words local contexts useful resolving language ambiguities improving retrieval output main text decomposition strategies introduced study including chronological decomposition text segments semantic decomposition text themes interaction text segments text themes used characterize text structure formulate specifications information retrieval text traversal text summarization

#* analysis syntax based pronoun resolution methods
#@Joel R. Tetreault
#t1999
#cACL
#index419
#id419
#%
#! paper presents pronoun resolution algo rithm adheres constraints rules centering theory grosz et al alternative brennan et al algorithm advantages new model left right centering algorithm lrc lie incremental processing utterances low computational overhead algorithm compared pronoun resolution methods hobbs syntax based algorithm strube list approach bfp

#* learning subjective adjectives corpora
#@Janyce Wiebe
#t2000
#cAAAI/IAAI
#index420
#id420
#%
#! subjectivity tagging distinguishing sentences used present opinions evaluations sentences used objectively present factual information numerous applications subjectivity tagging relevant including information extraction information retrieval paper identifies strong clues subjectivity using results method clustering words according distributional similarity lin seeded small detailed manual annotation

#* latent semantic indexing lsi trec report
#@S. Dumais
#t1994
#cTREC
#index421
#id421
#%
#! paper reports recent developments thelatent semantic indexing lsi retrieval method fortrec fi lsi uses reduced dimension vector spaceto represent words documentsfi importantaspect representation associationbetween terms automatically captured explicitlyrepresented used improve retrievalfiwe used lsi trec routing adhoctasksfi routing tasks lsi space wasconstructed using training documentsfi wecompared

#* topic based language models using em
#@Daniel Gildea,Thomas Hofmann
#t1999
#cProceedings of EUROSPEECH
#index422
#id422
#%
#! paper propose novel statistical language model capture topic related long range dependencies topics modeled latent variable framework derive em algorithm perform topic factor decomposition based segmented training corpus topic model combined standard language model used line word prediction perplexity results indicate improvement previously proposed topic models unfortunately translated lower word error

#* latent class models collaborative filtering
#@Thomas Hofmann,Jan Puzicha
#t1999
#cIJCAI
#index423
#id423
#%
#! paper presents statistical approach collaborative ltering investigates use latent class models predicting individ ual choices preferences based observed preference behavior models discussed compared aspect model proba bilistic latent space model models indi vidual preferences convex combination preference factors sided clustering model simultaneously partitions persons objects clusters present em algo rithms different variants aspect model derive approximate em algorithm based variational principle sided clus tering model bene ts different mod els experimentally investigated large movie data set

#* unsupervised learning dyadic data
#@Thomas Hofmann,Jan Puzicha
#t1999
#cAdvances in Neural Information Processing Systems
#index424
#id424
#%
#! dyadic data refers domain finite sets objects observa tions dyads pairs element set includes event occurrences histogram data single stimulus preference data special cases dyadic data arises naturally applications ranging computational linguistics information retrieval preference analysis vision paper present systematic domain independent framework unsupervised learning dyadic data statistical mixture models approach covers di er ent models hierarchical latent class structures unifies probabilistic modeling structure discovery mixture models provide parsimonious exible parameterization probability distributions good generalization perfor mance sparse data structural information data inherent grouping structure propose annealed version standard expectation maximization algorithm model fitting empirically evaluated variety data sets different domains

#* tdt pilot study corpus
#@
#t1000
#cEMPTY
#index425
#id425
#%
#! tdt corpus includes approximately stories half collected reuters newswire half cnn broadcast news transcripts period july june integral key corpus annotation terms news events discussed stories events defined span variety event types cover subset events discussed corpus stories annotation data events included corpus provide basis training tdt systems

#* mixture models
#@Petteri Nurmi
#t2004
#cnull
#index426
#id426
#%
#! mixture models interesting exible model family different uses mixture models include example generative component models clustering density estimation mixture models successfully used various kinds tasks modelling failure rate data clustering teaching behaviour general modelling large heterogeneous populations step using mixture models define suitable model data parameter model estimated data phase called parameter estimation parameter learning important task finding number subpopulations data supports problem called model learning paper discuss issues cover basic methods introduce modern methods numerous examples

#* view em algorithm justifies incremental variants
#@Radford Neal,Geoffrey E. Hinton
#t1998
#cLearning in Graphical Models
#index427
#id427
#%
#! em algorithm performs maximum likelihood estimation data variables unobserved present function resembles negative free energy step maximizes function respect model parameters step maximizes respect distribution unobserved variables perspective easy justify incremental variant em algorithm distribution unobserved variables

#* distributional clustering english words
#@Fernando C. N. Pereira,Naftali Tishby,Lillian Lee
#t1993
#cACL
#index428
#id428
#%
#! experimentally evaluate method automatically clustering words accordingto distribution particular syntactic contextsfi words represented relative frequencydistributions contexts appear relative entropy used measure dissimilarityof distributionsfi clusters represented typical context distributions averaged thegiven words according probabilities cluster membership cases

#* aggregate mixed order markov models statistical language processing
#@Lawrence K. Saul,Fernando Pereira
#t1997
#cProceedings of the Second Conference on Empirical Methods in Natural Language Processing
#index429
#id429
#%
#! consider use language models size accuracy intermediate different order gram models types models studied particular aggregate markov models classbased bigram models mapping words classes probabilistic mixed order markov models combine bigram models predictions conditioned different words types models trained expectationmaximization em algorithms maximum likelihood estimation examine smoothing procedures models interposed different order grams significantly reduce perplexity unseen word combinations

#* predicting performance linearly combined ir systems
#@Christopher C. Vogt,Garrison W. Cottrell
#t1998
#cSIGIR
#index430
#id430
#%
#! introduce new technique analyzingcombination modelsfi technique allows makequalitative conclusions ir systems shouldbe combinedfi achieve using linear regressionto accurately predict performance combined based quantitative measurementsof individual component systems taken trec fiwhen applied linear model weighted sum relevancescores technique supports previouslysuggested hypotheses

#* discovering objects location images
#@Josef Sivic, Bryan C. Russell,Alexei A. Efros,Andrew Zisserman, William T. Freeman
#t2005
#cInternational Conference on Computer Vision (ICCV 2005)
#index431
#id431
#%375,376,182,378,379,371,205,360,380,432,381,382,383,385,433,386,387,388,389
#! given set images containing multiple object categories seek discover categories image locations supervision achieve using generative models statistical text literature probabilistic latent semantic analysis plsa latent dirichlet allocation lda text analysis used discover topics corpus using bag words document representation discover topics object categories image containing instances categories modelled mixture topics models applied images using visual analogue word formed vector quantizing sift like region descriptors investigate set increasingly demanding scenarios starting image sets containing object categories sets containing multiple categories including airplanes cars faces motorbikes spotted cats background clutter object categories sample intra class scale variation categories approximate spatial layout supervision demonstrate classification unseen images images containing multiple objects performance proposed unsupervised method compared semisupervised approach

#* affine invariant point detector
#@Krystian Mikolajczyk,Cordelia Schmid
#t2002
#cECCV (1)
#index432
#id432
#%
#! paper presents novel approach detecting affine invariant pointsfi method deal significant affine transformations including large scale changesfi transformations introduce significant changes point location scale shape neighbourhood pointfi approach allows solve problems simultaneouslyfi based key ideas second moment matrix computed point used normalize

#* hierarchical dirichlet processes
#@Yee Whye Teh,Michael I. Jordan,Matthew J. Beal,David M. Blei
#t2006
#cJournal of the American Statistical Association
#index433
#id433
#%434,435
#! consider problems involving groups data observation group draw mixture model desirable share mixture components groups assume number mixture components unknown priori inferred data setting natural consider sets dirichlet processes group known clustering property dirichlet process provides nonparametric prior number mixture components group given desire tie mixture models various groups consider hierarchical model speci cally base measure child dirichlet processes itself distributed according dirichlet process base measure discrete child dirichlet processes necessar ily share atoms desired mixture models different groups necessarily share mixture components discuss representations hierarchical dirichlet processes terms stick breaking process generalization chinese restaurant process refer chinese restaurant franchise present markov chain monte carlo algorithms posterior inference hierarchical dirichlet process mixtures applications problems information retrieval text modelling

#* hierarchical bayesian models applications information retrieval
#@J. M. Bernardo,M. J. Bayarri,J. O. Berger,A. P. Dawid,D. Heckerman,A. F. M. Smith,M. West,David M. Blei,Michael I. Jordan,Andrew Y. Ng
#t2003
#cBayesian Statistics
#index434
#id434
#%
#! article factors close whilefour factors achieve significant expected countsfi looking distribution words factors identify topics mixed random variableto form document figure

#* hidden markov model induction bayesian model merging
#@A Stolcke,S Omohundro
#t1993
#cAdvances in Neural Information Processing Systems
#index435
#id435
#%572,573,574,575,576,577,578,579,580
#! paper describes technique learning number states topology hidden markov models examples induction process starts speci model consistent training data generalizes successively merging states choice states merge stopping criterion guided bayesian posterior probability compare algorithm baum welch method estimating xed size models nd induce minimal hmms data cases xed estimation does converge requires redundant parameters converge

#* dynamic social network analysis using latent space models
#@Purnamrita Sarkar,Andrew W. Moore
#t2005
#cNIPS
#index436
#id436
#%437,438,439,440,441
#! paper explores aspects social network modeling generalize successful static model relationships dynamic model accounts friendships drifting time second make tractable learn models data number entities gets large generalized model associates entity point dimensional euclidean latent space points time progresses large moves latent space improbable observed links entities likely entities close latent space make model tractable sub quadratic number entities use appropriate kernel functions similarity latent space use low dimensional kd trees new efficient dynamic adaptation multidimensional scaling pass approximate projection entities latent space efficient conjugate gradient update rule non linear local optimization amortized time entity update log use synthetic real world data entities indicate near linear scaling computation time improved performance alternative approaches illustrate operating years nips authorship data

#* crimelink explorer using domain knowledge facilitate automated crime association analysis
#@Jennifer Schroeder,Jennifer Jie Xu,Hsinchun Chen
#t2003
#cISI
#index437
#id437
#%
#! link association analysis used law enforcement intelligence domains extract search associations people large datasets nonetheless link analysis faces challenging problems information overload high search complexity heavy reliance domain knowledge address challenges enable crime investigators conduct automated effective efficient link analysis proposed techniques include concept space approach shortest path algorithm heuristic approach captures domain knowledge determining importance associations implemented called crimelink explorer based proposed techniques results user study involving crime investigators tucson police department showed help subjects conduct link analysis efficiently additionally subjects concluded association paths based heuristic approach accurate based concept space approach

#* clustering bipartite advertiser keyword graph
#@J. Carrasco,D. Fain,K. Lang,L. Zhukov
#t2003
#cProc. International Conference on Data Mining (ICDM'03)
#index438
#id438
#%
#! paper present hierarchicalclustering methods large bipartite graphsfithe approach employs flow based graph partitioningmethod approach multiroundhybrid single link average link agglomerativeclustering methodsfi evaluate quality clustersobtained methods using additional textual informationand compare results clusteringtechniquesfi

#* collaboration analysis recommender systems using social networks
#@Jordi Palau,Miquel Montaner,Beatriz Lopez,Josep Lluis de la Rosa
#t2004
#cCIA
#index439
#id439
#%
#! researchers focussed efforts developingcollaborative recommender systemsfi provedthat use collaboration systems improves itsperformance known collaborationis important tobe order optimise information exchange collaborative relationships recommender systems berepresented social networkfi paper proposeseveral measures analyse collaboration

#* modern multidimensional scaling
#@Ingwer Borg,Patrick Groenen
#t1996
#cbook
#index440
#id440
#%
#! book provides comprehensive treatment multidimensional scaling mds family statistical techniques analyzing structure dis similarity data data widespread including example intercorrelations survey items direct ratings similarity choice objects trade indices set countries mds represents data distances points geometric space low dimensionality map help patterns data obvious data matrices mds used psychological model judgments similarity preference book used introduction mds students psychology sociology marketing particular prerequisite elementary background statistics book suited variety advanced courses mds topics mathematics required advanced topics developed systematically

#* optimization em expectation conjugategradient
#@Ruslan Salakhutdinov,Sam Roweis,Zoubin Ghahramani
#t2003
#cICML 2003
#index441
#id441
#%
#! close relationship expectation maximization em algorithm direct optimization algorithms gradientbased methods parameter learning identify analytic conditions em exhibits newton like behavior conditions possesses poor order convergence

#* dynamic topic models
#@David M. Blei,J Lafferty
#t2006
#cICML
#index442
#id442
#%182,346,371,443,351,444,431,445,446,447
#! family probabilistic time series models developed analyze time evolution topics large document collections approach use state space models natural parameters multinomial distributions represent topics variational approximations based kalman filters nonparametric wavelet regression developed carry approximate posterior inference latent topics addition giving quantitative predictive models sequential corpus dynamic topic models provide qualitative window contents large document collection models demonstrated analyzing ocred archives journal science

#* new approach linear filtering prediction problems
#@R. E. Kalman
#t1960
#cASME
#index443
#id443
#%
#! classical filtering prediction problem examined using bodeshannon representation random processes state transition method analysis dynamic systems new results formulation methods solution problem apply modification stationary nonstationary statistics growing memory infinitememory filters nonlinear difference differential equation derived covariance matrix optimal estimation error solution equation coefficients difference differential equation optimal linear filter obtained calculations filtering problem shown dual noise free regulator problem new method developed applied known problems confirming extending earlier results discussion largely self contained proceeds principles basic concepts theory random processes reviewed appendix

#* author topic model authors documents
#@Michal Rosen-Zvi,Thomas L. Griffiths,Mark Steyvers,Padhraic Smyth
#t2004
#cUAI
#index444
#id444
#%182,456,372,371,205,457,458,452,211
#! introduce author topic model generative model documents extends latent dirichlet allocation lda blei ng jordan include authorship information author associated multinomial distribution topics topic associated multinomial distribution words document multiple authors modeled distribution topics mixture distributions associated authors apply model collection nips conference papers citeseer exact inference intractable datasets use gibbs sampling estimate topic author distributions compare performance generative models documents special cases author topic model lda topic model simple author model author associated distribution words distribution topics topics recovered authortopic model demonstrate applications computing similarity authors entropy author output

#* sparse gaussian processes using pseudo inputs
#@Edward Snelson,Zoubin Ghahramani
#t2005
#cNIPS
#index445
#id445
#%
#! present new gaussian process gp regression model covarianceis parameterized locations pseudo input points learn gradient based optimizationfi mn number real data points obtain sparseregression method hasn training cost predictioncost test casefi hyperparameters covariancefunction joint optimizationfi method viewedas bayesian

#* nonparametric statistics
#@L Wasserman
#t2006
#cbook
#index446
#id446
#%
#! text provides reader single book accounts number date issues nonparametric inference book aimed masters phd level students statistics science engineering suitable researchers want speed quickly modern nonparametric methods covers wide range topics including bootstrap nonparametric delta method nonparametric regression density estimation orthogonal function methods minimax estimation nonparametric confidence sets wavelets books dual approach includes mixture methodology theory

#* bayesian forecasting dynamic models
#@M. West,Jeff Harrison
#t1997
#cbook
#index447
#id447
#%
#! textbook second edition book includes revised updated additional material structure theory application classes dynamic models bayesian time series analysis forecasting addition wide ranging updates central material second edition includes exercises covers new topics research application frontiers bayesian forecastings

#* author recipient topic model topic role discovery social networks experiments enron academic email
#@A McCallum,X Wang,A Corrada-Emmanuel
#t2007
#cJournal of Artificial Intelligence Research
#index448
#id448
#%449,182,450,451,452,211,453,433,454,455
#! previous work social network analysis sna modeled existence links entity language content topics links present author recipient topic art model social network analysis learns topic distributions based directionsensitive messages sent entities model builds latent dirichlet allocation author topic model adding key attribute distribution topics conditioned distinctly sender recipientsteering discovery topics according relationships people results enron email corpus researchers email archive providing evidence clearly relevant topics discovered art model better predicts peoples roles

#* search social network
#@LA Adamic,Eytan Adar
#t2003
#cbook
#index449
#id449
#%
#! address question participants small world experiment able short paths social network using local information immediate contactsfi simulate experiments network actual email contacts organization student social networking websitefi email network small world search strategies using contact position physical space organizational hierarchy relative target efiectively used locate individualsfi line student network data incomplete hierarchical structures defined local search strategies efiectivefi compare findings recent theoretical hypotheses underlying social structure en able simple search strategies succeed discuss implications social software designfi

#* navigation small world
#@Jon M. Kleinberg
#t2000
#cnature
#index450
#id450
#%
#! small world phenomenon principle linked short chains acquaintances investigated question sociology feature range networks arising nature technology experimental study phenomenon revealed fundamental components short chains ubiquitous second individuals operating purely local information adept finding chains issue analysed investigate second modelling individuals short chains large social network page

#* structural equivalence individuals social networks
#@F. lorrain,H. C. white
#t1971
#cJournal of Mathematical Sociology
#index451
#id451
#%
#! problems structural sociology attracted attention researchers creation individual identity means social networks individuals embedded second structural equivalence problem non identification individuals means social networks aim article bring problems common framework identity equivalence individuals social networks simultaneously emerge words study attempts structural determinations identity equivalence just order achieve necessary duality approach insofar duality represents social structure network interconnected social circles network individuals linked given relations accordingly concept place place analysis used

#* multi label text classification mixture model trained em
#@A McCallum
#t1999
#cAAAI 99 Workshop on Text Learning
#index452
#id452
#%
#! important document classification tasks documents associated multiple class labelsfi paper describes bayesian classification approach multiple classes comprise document represented mixture modelfi labeled training data indicates classes responsible generating document does indicate class responsible generating wordfi use em missing value learning

#* probabilistic author topic models information discovery
#@Mark Steyvers,Padhraic Smyth,Michal Rosen-Zvi,Thomas L. Griffiths
#t2004
#cKDD
#index453
#id453
#%
#! propose new unsupervised learning technique extracting information large text collectionsfi model documents generated stage stochastic processfi author represented probability distribution topics topic represented probability distribution words topicfi words multi author paper assumed result mixture authors topic mixturefi topic word author topic distributions learned data unsupervised manner using markov chain monte carlo algorithmfi apply methodology large corpus authors known citeseer digital library learn model topicsfi discuss interpretation results discovered including specific topic author models ranking authors topic topics author significant trends science literature parsing abstracts topics authors detection unusual papers specific authorsfi online query interface model discussed allows interactive exploration author topic models corpora citeseerfi

#* identify search social networks
#@D. J. Watts,P. S. Dodds,M. E. J. Newman
#t2002
#cScience
#index454
#id454
#%
#! social networks surprising property searchable ordinary people capable directing messages network acquaintances reach specific distant target person steps present model ers explanation social network searchability terms recognizable personal identities defined number social dimensions model defines class searchable networks method searching applicable network search problems including location data files peer peer networks pages world wide web information distributed databases

#* information flow social groups
#@F Wu,BA Huberman,LA Adamic,JR Tyler
#t2004
#cPhysica A: Statistical Mechanics and its Applications
#index455
#id455
#%
#! present study information ow takes account observation item relevant person likely individuals social circle outside fact similarity node attributes social networks decreases function graph distance epidemic model scale free network property finite threshold implying spread information limited tested predictions measuring spread messages organization numerical experiments consideration organizational distance individuals

#* missing link probabilistic model document content hypertext connectivity
#@David Cohn,Thomas Hofmann
#t2001
#cNeural Information Processing Systems 13
#index456
#id456
#%
#! joint probabilistic model modeling contents inter connectivity document collections sets web pages research paper archives model based probabilistic factor decomposition allows identifying principal topics collection authoritative documents topics furthermore relationships topics mapped order build predictive model link content applications approach information retrieval search topic identification query disambiguation focused web crawling web authoring bibliometric analysis

#* federalist revisited new directions authorship attribution
#@D. I. HOLMES,R. S. FORSYTH
#t1995
#cLiterary and Linguistic Computing
#index457
#id457
#%
#! federalist papers claimed alexander hamilton james madison long used testing ground authorship attribution techniques despite fact styles hamilton madison unusually similar paper assesses value novel stylometric techniques applying federalist problem techniques examined multivariate approach vocabulary richness analysis frequencies occurrence sets common high frequency words use machine learning package based genetic algorithm seek relational expressions characterizing authorial styles approaches produce encouraging results acknowledged difficult problem

#* digital libraries autonomous citation indexing
#@S Lawrence,C. Lee Giles,Kurt D. Bollacker
#t1999
#cIEEE Computer
#index458
#id458
#%
#! world wide web revolutionizing way researchers access scientific informationfi articles increasingly available homepages authors institutions journal web sites online archivesfi scientific information web largely disorganizedfi article introduces creation digital libraries incorporating autonomous citation indexing aci fi aci autonomously creates citation indices similar science citation index fi aci

#* efficient distributed skylining web information systems
#@Wolf-Tilo Balke,Ulrich Guntzer,Jason Xin Zheng
#t2004
#cEDBT
#index459
#id459
#%460,461,232,462,238,463,243,242,464,465,466,467,468,469
#! skyline queries claimed place retrieval central databases application web information systems impossible distributed aspect retrieval web sources variety volatile nature information accessible internet extended query capabilities crucial efficiently perform distributed skyline queries essentially extend expressiveness querying todays web information systems innovative retrieval algorithm present useful heuristics speed retrieval practical cases paving road meeting realtime challenges line information services discuss performance evaluations point open problems concept application skylining modern information systems curse dimensionality intrinsic problem skyline queries propose novel sampling scheme allows early impression skyline subsequent query refinement

#* situation aware mobile traffic information prototype
#@WT Balke,W Kiessling,C Unbehend
#t2003
#cSystem Sciences
#index460
#id460
#%
#! mobile information services play important role future work private life enabling mobility urban populous areas needs innovative tools novel techniques individual traffic planning navigation systems featuring route planning usability difficult current information personal user preferences incorporated present prototype traffic information offering advanced personalized route planning including added services like traffic jam alerting means sms fast changing nature data requires gather line internet sources given current bandwidth limitations asynchronous update strategy central service database prepares ground meet real time requirements providing todate information results personalized route planning query efficiently computed srcombine algorithm seconds practical case studies xslt technology automatically converts results delivery various mobile devices summary evidence intelligently joining latest technologies preference modeling preference query evaluation advanced personalized mobile traffic information systems feasible today

#* maximin fitness function multi objective city regional planning
#@BALLING Richard
#t2003
#c2003	International conference on evolutionary multi-criterion optimization
#index461
#id461
#%
#! maximin fitness function used multi objective genetic algorithms obtain diverse set non dominated designs maximin fitness function derived definition dominance properties explored modified maximin fitness function proposed fitness functions briefly compared state art fitness function literature results real world multi objective problem presented problem addresses land use transportation planning high growth cities metropolitan regions

#* querying intrinsic preferences
#@Jan Chomicki
#t2002
#cEDBT
#index462
#id462
#%
#! handling user preferences increasingly important issue present day information systems preferences used information filtering extraction reduce volume data presented user used track user profiles formulate policies improve automate decision making propose logical framework formulating preferences embedding relational query languages framework simple entirely neutral respect properties preferences makes possible formulate different kinds preferences use preferences querying databases demonstrate usefulness framework numerous examples

#* optimizing multi feature queries image databases
#@Ulrich Guntzer,Wolf-Tilo Balke,Werner Kiesling
#t2000
#cVLDB
#index463
#id463
#%
#! digital libraries image retrieval queries bebased similarity objects using severalfeature attributes like shape texture color textfi

#* shooting stars sky online algorithm skyline queries
#@Donald Kossmann,Frank Ramsak,Steffen Rost
#t2002
#cVLDB
#index464
#id464
#%
#! skyline queries ask set interesting points potentially large set data pointsfi traveling instance restaurant interesting restaurant nearer cheaper better foodfi skyline queries retrieve interesting restaurants user choose promising onefi paper present new online algorithm computes skylinefi unlike existing algorithms compute skyline batch

#* preferences putting knowledge queries
#@M Lacroix,P Lavency
#t1987
#cProceedings of the 13th VLDB Conference
#index465
#id465
#%
#! classical query languages provide way express mandatory qualifications data retrieved feature facilities expressing preferences desirable qualifications need preferences illustrated software engineering framework preference mechanism presented extension language domain relational calculus family expressive power resulting language discussed proposed mechanisms shown effectively allow use queries supporting software configuration management functions

#* efficient utility functions ceteris paribus preferences
#@Michael McGeachie,Jon Doyle
#t2002
#cAAAI/IAAI
#index466
#id466
#%
#! ceteris paribus things equal preference provides convenient means stating constraints numeric utility functions direct constructions numerical utility representations statements exponential worstcase cost paper describes efficient constructions combine analysis utility independence constraintbased search

#* supporting ranked boolean similarity queries mars
#@Michael Ortega,Yong Rui,Kaushik Chakrabarti,Kriengkrai Porkaew,Sharad Mehrotra,T.Huang
#t1998
#cIEEE Trans. Knowl. Data Eng.
#index467
#id467
#%
#! address emerging needs applications require access retrieval multimedia objects weare developing multimedia analysis retrieval mars fi paper concentrate onthe retrieval subsystem mars support content based queries image database content basedretrieval techniques extensively studied textual documents area automatic informationretrieval fi paper describes techniques

#* optimal progressive algorithm skyline queries
#@Dimitris Papadias,Yufei Tao,Greg Fu,Bernhard Seeger
#t2003
#cSIGMOD Conference
#index468
#id468
#%
#! skyline set dimensional points contains points dominated point dimensions skyline computation recently received considerable attention database community especially progressive online algorithms quickly return skyline points having read entire data file currently efficient algorithm nn nearest neighbors applies divide conquer framework datasets indexed trees nn desirable features high speed returning initial skyline points applicability arbitrary data distributions dimensions presents inherent disadvantages need duplicate elimination multiple accesses node large space overhead paper develop bbs branch bound skyline progressive algorithm based nearest neighbor search io optimal performs single access tree nodes contain skyline points furthermore does retrieve duplicates space overhead significantly smaller nn finally bbs simple implement efficiently applied variety alternative skyline queries analytical experimental comparison shows bbs outperforms nn usually orders magnitude problem instances

#* roadmap advanced personalization mobile services
#@M. Wagner,WT Balke,M. Wagner,Wolf-Tilo Balke,R. Hirschfeld,W. Kellerer
#t2002
#c10th International Conference on Cooperative Information Systems
#index469
#id469
#%
#! performing complex tasks web integral everyday lifefi

#* efficient document retrieval main memory
#@Trevor Strohman,W. Bruce Croft
#t2007
#cSIGIR
#index470
#id470
#%471,472,473,38,474,43,475,238,476,46,47,477,478
#! disk access performance major bottleneck traditional information retrieval systems compared memory disk bandwidth poor seek times worse circumvent problem considering query evaluation strategies main memory new accumulator trimming techniques combined inverted list skipping produce extremely high performance retrieval systems resorting methods harm effectiveness evaluate techniques using galago new retrieval designed efficient query processing achieves improvement query throughput previous methods

#* vector space ranking effective early termination
#@Vo Ngoc Anh,Owen de Kretser,Alistair Moffat
#t2001
#cSIGIR
#index471
#id471
#%
#! considerable research effort invested improving effectiveness information retrieval systems techniques relevance feedback thesaural expansion pivoting provide better quality responses queries tested standard evaluation frameworks enhancements add cost evaluating queries paper consider pragmatic issue improve cost effectiveness searching new inverted file structure using quantized weights provides superior retrieval effectiveness compared conventional inverted file structures early termination heuristics employed able reach similar effectiveness levels computational cost provide better cost performance compromise previous inverted file organisations

#* simplified similarity scoring using term ranks
#@Vo Ngoc Anh,Alistair Moffat
#t2005
#cSIGIR
#index472
#id472
#%
#! propose method document ranking combines simple document centric view text fast evaluation strategies developed connection vector space model new method defines importance term document qualitatively quantitatively doing reduces need tuning parameters addition method supports fast query processing computation carried small integers dynamic pruning effective option experiments wide range trec data new method provides retrieval effectiveness good better okapi bm formulation variants language models

#* pruned query evaluation using pre computed impacts
#@Vo Ngoc Anh,Alistair Moffat
#t2006
#cSIGIR
#index473
#id473
#%
#! exhaustive evaluation ranked queries expensive particularly small subset overall ranking required queries contain common terms concern gives rise techniques dynamic query pruning methods eliminating redundant parts usual exhaustive evaluation generating demonstrably good set answers query work propose new pruning methods make use impact sorted indexes compared exhaustive evaluation new methods reduce computation performed reduce memory required accumulators reduce data transferred disk time allow performance guarantees terms precision mean average precision strong claims backed experiments using trec terabyte collection queries

#* trec terabyte track
#@Stefan Buttcher,CLA Clarke,Ian Soboroff
#t2006
#cTREC
#index474
#id474
#%
#! trec year track track introduced trec single adhoc retrieval task trec track expanded optional tasks named page finding task efficiency task tasks continued groups submitting runs adhoc retrieval task groups submitting runs named page finding task groups submitting runs efficiency task report provides overview task summarizes results outlines directions future background information development track track reports

#* mapreduce simplified data processing large clusters
#@Jeffrey Dean,Sanjay Ghemawat
#t2008
#cCommun. ACM
#index475
#id475
#%479,480,481,482,483,484,485,486,487,488,489,490,491,492,493
#! mapreduce programming model associated implementation processing generating large data sets users specify map function processes key value pair generate set intermediate key value pairs reduce function merges intermediate values associated intermediate key real world tasks expressible model shown paper programs written functional style automatically parallelized executed large cluster commodity machines run time takes care details partitioning input data scheduling program execution set machines handling machine failures managing required inter machine communication allows programmers experience parallel distributed systems easily utilize resources large distributed implementation mapreduce runs large cluster commodity machines highly scalable typical mapreduce computation processes terabytes data thousands machines programmers easy use hundreds mapreduce programs implemented upwards thousand mapreduce jobs executed google clusters day

#* space limited ranked query evaluation using adaptive pruning
#@Nicholas Lester,Alistair Moffat,William Webber,Justin Zobel
#t2005
#cWISE
#index476
#id476
#%
#! evaluation ranked queries large text collections costly terms processing time memory space dynamic pruning techniques allow costs reduced potential risk decreased retrieval effectiveness paper improved query pruning mechanism offers resilient tradeoff query evaluation costs retrieval effectiveness previous pruning approaches

#* optimization strategies complex queries
#@Trevor Strohman,Howard R. Turtle,W. Bruce Croft
#t2005
#cSIGIR
#index477
#id477
#%494,64,495,496,46,62
#! previous research efficiency text retrieval systems dealt primarily methods consider inverted lists sequence methods known term time methods literature optimizing documentata time systems remains sparse present improvement max score optimization efficient known document time scoring method like max score technique called term bounded max score guaranteed return exactly scores documents unoptimized evaluation particularly useful query model research simulated technique explore problem space implemented indri large scale language modeling search engine tests gov corpus title queries method faster max score faster document time baseline optimized query times competitive conventional termata time systems years trec terabyte task

#* monetdb dbms cpu cache
#@Marcin Zukowski,Peter A. Boncz,Niels Nes,Sandor Heman
#t2005
#cIEEE Data Eng. Bull.
#index478
#id478
#%
#! new execution engine monetdb improves execution speed overcomes main memory limitation introduces concept cache vectorized processing strikes balance existing column time mil execution primitives monetdb tuple time volcano pipelining model avoiding drawbacks intermediate result materialization large interpretation overhead respectively explain new query engine makes better use cache memories parallel computation resources modern super scalar cpus monetdb orders magnitude faster commercial dbmss close hand coded programs computationally intensive queries memory datasets address larger disk based datasets efficiency new columnbm storage layer developed boosts bandwidth using ultra lightweight compression cooperative scans

#* high performance sorting networks workstations
#@Andrea C. Arpaci-Dusseau,Remzi H. Arpaci-Dusseau,David E. Culler,Joseph M. Hellerstein,David A. Patterson
#t1997
#cSIGMOD Conference
#index479
#id479
#%
#! report performance sort collection sorting implementations network workstations parallel sorting competitive sorting large scale smps traditionally held performance records node cluster sort gb just minute node cluster finishes datamation benchmark seconds implementations applied variety disk memory processor configurations highlight salient issues tuning component evaluate use commodity operating systems hardware parallel sorting existing os primitives memory management file access adequate aggregate communication disk bandwidth requirements bottleneck workstation bus

#* cluster river making fast case common
#@Remzi H. Arpaci-Dusseau,Eric Anderson,Noah Treuhaft,David E. Culler,Joseph M. Hellerstein,David A. Patterson,Katherine A. Yelick
#t1999
#cIOPADS
#index480
#id480
#%
#! introduce river data flow programming environment substrate clusters computers river designed provide maximum performance common case face non uniformities hardware software workload river based simple design features high performance distributed queue storage redundancy mechanism called graduated declustering implemented number data intensive applications river validate design

#* charlotte metacomputing web
#@Arash Baratloo,Mehmet Karaul,Zvi M. Kedem,P. Wijckoff
#t1999
#cFuture Generation Comp. Syst.
#index481
#id481
#%
#! world wide web potential used inexpensive convenient metacomputing resource brings forward new challenges invalidates assumptions offering functionality network workstations designed implemented charlotte goes providinga set features commonly used network workstations user execute parallel program machine does account shared

#* web search planet google cluster architecture
#@Luiz Andre Barroso,Jeffrey Dean,Urs Holzle
#t2003
#cIEEE Micro
#index482
#id482
#%
#! application lets different queries run different processors partitioning overall index lets single query use multiple processors handle workload googles architecture features clusters commodityclass pcs fault tolerant software architecture achieves superior performance fraction cost built fewer expensive high end servers

#* explicit control batch aware distributed file
#@John Bent,Douglas Thain,Andrea C. Arpaci-Dusseau,Remzi H. Arpaci-Dusseau,Miron Livny
#t2004
#cNSDI
#index483
#id483
#%
#! present design implementation evaluation batch aware distributed file bad fs designed orchestrate large intensive batch workloads remote computing clusters distributed wide area bad fs consists novel components storage layer exposes control traditionally fixed policies caching consistency replication scheduler exploits control necessary different workloads extracting control storage layer placing external scheduler bad fs manages storage computation coordinated way gracefully dealing cache consistency fault tolerance space management issues workload specific manner using microbenchmarks real workloads demonstrate performance benefits explicit control delivering excellent end end performance wide area

#* scans primitive parallel operations
#@Guy E. Blelloch
#t1989
#cIEEE Trans. Computers
#index484
#id484
#%
#! study effects adding scan primitives unit time primitives pram parallel random access machine models presented shown primitives improve asymptotic running time algorithms log factor greatly simplifying description algorithms significantly easier implement memory references argued algorithm designer feel free use operations cheap memory reference author describes algorithms clearly illustrate scan primitives used algorithm design radix sort algorithm quicksort algorithm minimum spanning tree algorithm line drawing algorithm merging algorithm run erew exclusive read exclusive write pram addition scan primitives simpler efficient pure pram counterparts scan primitives implemented microcode connection machine available paris parallel instruction set machine

#* cluster based scalable network services
#@Armando Fox,Steven D. Gribble,Yatin Chawathe,Eric A. Brewer,Paul Gauthier
#t1997
#cSOSP
#index485
#id485
#%
#! identify fundamental requirements scalable net work services incremental scalability overflow growth provi sioning availability fault masking cost effectiveness argue clusters commodity workstationsinterconnected high speed san exceptionally suitedto meeting challenges internet server workloads pro vided software infrastructure managing partial failures andadministering large cluster does reinvented foreach new service end propose general layered archi tecture building cluster based scalable network services thatencapsulates requirements reuse service pro gramming model based composable workers perform trans formation aggregation caching customization tacc ofinternet content performance implementation sim plicity architecture tacc programming model exploitbase weaker acid data semantics results trad ing consistency availability relying soft state robust ness failure management architecture used anoff shelf infrastructural platform creating new networkservices allowing authors focus content service composing tacc building blocks implementa tion discuss real implementations services based thisarchitecture transend web distillation proxy deployed theuc berkeley dialup ip population hotbot commercialimplementation inktomi search engine present detailedmeasurements transends performance based substantial cli ent traces anecdotal evidence transend andhotbot experience support claims architecture

#* systematic efficient parallelization scan list homomorphisms
#@Sergei Gorlatch
#t1996
#cEuro-Par, Vol. II
#index486
#id486
#%
#! homomorphisms functions parallelized divide conquer paradigmfi class distributable homomorphisms dh introduced efficient parallel implementation schema functions class derived transformations bird meertens formalismfi schema directly mapped hypercube unlimited arbitrary fixed number processors providing provable correctness predictable performancefi popular scan function parallel

#* diamond storage architecture early discard interactive search
#@Larry Huston,Rahul Sukthankar,Rajiv Wickremesinghe,Mahadev Satyanarayanan,Gregory R. Ganger,Erik Riedel,Anastassia Ailamaki
#t2004
#cFAST
#index487
#id487
#%
#! paper explores concept early discard teractive search unindexed data processing data storage devices using downloaded searchlet code enables diamond perform efficient application specific filtering large data collections early dis card helps users looking needles haystack eliminating bulk irrelevant items early possible searchlet consists set application generated filters diamond uses deter object user optimizes evaluation order filters based run time measurements filter selec tivity computational cost diamond dy namically partition computation storage vices host adjust changes hardware network conditions performance num bers diamond dynamically adapts query run time state informal user study image retrieval application supports belief early discard significantly improves quality inter active searches

#* parallel prefix computation
#@Richard E. Ladner,Michael J. Fischer
#t1980
#cJ. ACM
#index488
#id488
#%
#! prefix problem compute products fi xk kfi associative operation recurstve construction used obtain product circuit solving prefix problem depth exactly size bounded application yields fast small boolean ctrcmts simulate fimte state transducersfi simulating sequentml adder boolean clrcmt depth size bounded obtained bit binary addmon size decreased significantly permitting depth increase addmve constant

#* efficient dispersal information security load balancing fault tolerance
#@Michael Rabin
#t1989
#cJournal of the ACM
#index489
#id489
#%
#! information dispersal algorithm ida developed breaks file length uharl uharr pieces fi length uharl fi uharr pieces suffice reconstructing dispersal reconstruction computationally efficient sum lengths uharl fi uharr chosen close ida space efficient ida numerous applications secure reliable storage information networks single disks fault tolerant efficient transmission information networks communications processors parallel computers problem provably time efficient highly fault tolerant routing cube achieved using just constant size buffers

#* active disks large scale data processing
#@Erik Riedel,Christos Faloutsos,Garth A. Gibson,David Nagle
#t2001
#cIEEE Computer
#index490
#id490
#%
#! ory cost decreases intelligence tinues away cpu peripheralsfi storage designers use trend excess computing power form complex processing optimizations storage devicesfi date optimizations place relatively low levels storage protocolfi trends storage density mechanics electronics eliminate hardware bottleneck pressure interconnects hosts data efficientlyfi propose using active disk storage device combines drive processing memory soft ware downloadability allow disks execute appli cation level functions directly devicefi moving portions application processing storage device significantly reduces data traffic leverages parallelism present large systems dra matically reducing execution time basic data mining tasksfi

#* distributed computing practice condor experience
#@Douglas Thain,Todd Tannenbaum,Miron Livny
#t2005
#cConcurrency - Practice and Experience
#index491
#id491
#%
#! condor project enabled ordinary users extraordinary computing today project continues explore social technical problems cooperative computing scales ranging desktop world wide computational grid chapter provide history philosophy condor project interacted projects evolved field distributed computing outline core components condor technology computing correspond social structures reflect lessons experience chart course traveled research ideas grow production systems

#* bridging model parallel computation
#@Leslie G. Valiant
#t1990
#cCommun. ACM
#index492
#id492
#%
#! success von neumann model sequential computation attributable fact efficient bridge software hardware high level languages efficiently compiled model effeciently implemented hardware author argues analogous bridge software hardware required parallel computation widely used article introduces bulk synchronous parallel bsp model candidate role gives results quantifying efficiency implementing high level language features algorithms implemented hardware

#* spsort sort terabyte quickly
#@Jim Wyllie
#t1999
#cTechnical Report,IBM Almaden Lab
#index493
#id493
#%
#! december node ibm rs sp sorted terabyte data billion byte records minutes secondsfi fi times faster previous record problem thismagnitudefi spsort program itself custom designed benchmark cluster itsinterconnection hardware disk subsystem operating file communication library jobmanagement software ibm productsfi sustained aggregate data

#* efficient query evaluation using level retrieval process
#@AZ Broder,David Carmel,Michael Herscovici,Aya Soffer,Jason Y. Zien
#t2003
#cCIKM
#index494
#id494
#%497,498,499,500,64,501,502,503,504,67,505,68,69,506,507
#! present efficient query evaluation method based level approach level method iterates parallel query term postings identifies candidate documents using approximate evaluation taking account partial information term occurrences query independent factors second level promising candidates fully evaluated exact scores computed efficiency evaluation process improved significantly using dynamic pruning techniques little cost effectiveness pruning controlled user function time allocated query evaluation experimentally using trec web track data determined algorithm significantly reduces total number evaluations loss precision recall heart approach efficient implementation new boolean construct called wand weak independent

#* inquery retrieval
#@James P. Callan,W. Bruce Croft,Stephen M. Harding
#t1992
#cDEXA
#index495
#id495
#%
#! larger heterogeneous text databases available information retrieval research depend development powerful efficient flexible retrieval engines paper retrieval inquery based probabilistic retrieval model provides support sophisticated indexing complex query formulation inquery used successfully databases containing nearly documents introduction increasing

#* combining language model inference network approaches retrieval
#@Donald Metzler,W. Bruce Croft
#t2004
#cInf. Process. Manage.
#index496
#id496
#%
#! inference network retrieval model implemented inquery search engine allows richly structured queries incorporates form ad hoc tf idf estimates word probabilities language modeling offers formal estimation techniques paper combine language modeling inference network approaches single framework resulting model allows structured queries evaluated using language modeling estimates explore issues involved combining beliefs smoothing proximity nodes experimental results presented comparing query likelihood model inquery new model results reaffirm high quality structured queries outperform unstructured queries consistently achieves higher average precision inquery

#* fast match continuous speech recognition using allophonic models
#@L.R. Bahl, Peter V. de Souza, P.S. Gopalakrishnan, D. Nahamoo,M.A. Picheny 
#t1992
#c1992 IEEE ICASSP
#index497
#id497
#%
#! large vocabulary real time speech recognition need fast method selecting list candidate words vocabulary match given acoustic input paper highly accurate fast acoustic match continuous speech recognition algorithm uses allophonic models efficient search techniques select set candidate words allophonic models derived constructing decision trees query context phone occurs arrive allophone given context models words vocabulary arranged tree structiire efficient tree search algorithms used select list candidate words using models using method able obtain accuracy fast makh continuous speech recognition task vocabulary words

#* pivot join runtime operator text search
#@K. Beyer,A. Jhingran,B. Lyle,S. Rajagopalan,E. Shekita
#t2003
#cnull
#index498
#id498
#%
#! consider problem efficiently sampling web search engine query results turn using small random sample instead set results leads efficient approximate algorithms applications determining set categories given taxonomy spanned search results finding range metadata values associated result set order enable multi faceted search estimating size result set data mining associations query terms present analyze efficient algorithm obtaining uniform random samples applicable search engine based posting lists document time evaluation knowledge popular web search engines google inktomi altavista alltheweb belong class furthermore algorithm modified follow modern object oriented approach posting lists viewed streams equipped method method boolean complex queries built method primitive terms case construct basic method samples term posting lists probability construct methods boolean operators wand primitive methods finally test efficiency quality approach synthetic real world data

#* space time trade offs hash coding allowable errors
#@Burton H. Bloom
#t1970
#cCommun. ACM
#index499
#id499
#%
#! paper trade offs certain computational factors hash coding analyzed paradigm problem considered testing series messages membership given set messages new hash coding methods examined compared particular conventional hash coding method computational factors considered size hash area space time required identify message nonmember given set reject time allowable error frequency new methods intended reduce space required contain hash coded information associated conventional methods reduction space accomplished exploiting possibility small fraction errors commission tolerable applications particular applications large data involved core resident hash area consequently feasible using conventional methods applications envisaged overall performance improved using smaller core resident hash area conjunction new methods necessary using secondary time consuming test catch small fraction errors associated new methods example discussed illustrates possible areas application new methods analysis paradigm problem demonstrates allowing small number test messages falsely identified members given set permit smaller hash area used increasing reject time

#* anatomy large scale hypertextual web search engine
#@Sergey Brin,Lawrence Page
#t1998
#cComputer Networks
#index500
#id500
#%
#! paper present google prototype large scale search engine makes heavy use structure present hypertext google designed crawl index web efficiently produce satisfying search results existing systems prototype text hyperlink database million pages available http oogle stanford edu engineer search engine challenging task search engines index tens hundreds millions web pages involving comparable number distinct terms answer tens tillions queries day despite importance large scale search engines web little academic research furthermore rapid advance technology web proliferation creating web search engine today different years ago paper provides depth description large scale web search engine detailed public description know date apart problems scaling traditional searchtechniques data magnitude new technicalchallenges involved using additional information present hypertext produce better search results paper addresses question build practical large scale exploit additional infon nation present hypertext look problem effectively deal uncontrolled hypertext collections publish want

#* object oriented interface index
#@Burrows,Michael
#t2000
#cpatent
#index501
#id501
#%
#! implemented method searches index locate records database using object oriented interface record unique address database index organized plurality index entries index entry including word ordered list locations word occurs database words represent unique piece information database index entries ordered according collating order words second according collating order locations associated word query parsed terms operators term associated corresponding index entry operators relate terms basic stream reader object generated term query basic stream reader object sequentially reads locations corresponding index entry determine target location compound stream reader object generated operator compound stream reader object references basic stream reader objects associated terms related operator compound stream reader object produces locations words single record according operator basic compound stream reader object encapsulation data references method references operate data references

#* method parsing indexing searching world wide web pages
#@Patent
#t1000
#cEMPTY
#index502
#id502
#%
#! indexes web pages internet pages stored computers distributively connected communications network page unique url universal record locator pages include url links pages communication interface connected internet used fetching batch web pages computers accordance urls url links urls determined automated web browser connected communications interface parser sequentially partitions batch specified pages indexable words word represents indexable portion information specific page word represents attribute portions specific page parser sequentially assigns locations words parsed locations indicates unique occurrences word web output parser stored memory index index includes index entry unique word index entry includes location entries indicating unique word occurs web query module parses query terms operators operators relate terms search engine uses object oriented stream readers sequentially read location specified index entries specified index entries correspond terms query display module presents qualified pages located search engine users web

#* juru trec experiments index pruning
#@David Carmel,Einat Amitay,Miki Herscovici,Yoelle S. Maarek,Yael Petruschka,Aya Soffer
#t2001
#cTREC
#index503
#id503
#%
#! year juru java ir developed past years ibm research lab haifa participated trecs web track experiments focused ad hoc tasks main goal experiments validate novel pruning method presented significantly reduces size index little influence systems precision reducing index size feasible index large text collections web tracks data low end machines furthermore using method web search engines significantly decrease burden storing backing extremely large indices discarding entries influence search results

#* shortest substring ranking
#@CLA Clarke,Gordon V. Cormack,Forbes J. Burkowski
#t1995
#cProceedings of TREC-4
#index504
#id504
#%
#! address trec topics used precise query language yields combines arbitrary intervals text pre defined units like words documents solution scored inverse proportion length shortest interval containing document scored sum scores solutions strategy yielded documents documents satisfying successively weaker queries added lower rank results

#* term ordered query evaluation versus document ordered query evaluation large document databases
#@Marcin Kaszkiel,Justin Zobel
#t1998
#cSIGIR
#index505
#id505
#%
#! main families technique eficient processing ranked queries large text collections document ordered processing term ordered processing note compare techniques experimentally similar costs short queries long queries document ordered processing costly overall conclude term ordered processing refinements limited accumulators hierarchical index structuring eficient mechanism

#* optimal semijoins distributed database systems
#@James K. Mullin
#t1990
#cIEEE Trans. Software Eng.
#index506
#id506
#%
#! bloom filter based semijoin algorithm distributed database systems presentedfi algorithm reduces communications costs process distributed natural join possible filter approachfi optimal filter developed piecesfi filter information used recognize semijoin cease effective optimally process semijoinfi ineffective semijoin quickly cheaply recognizedfi effective semijoin use transmitted bits optimallyfi

#* overview tenth text retrieval conference trec
#@EM Voorhees,Donna Harman
#t2001
#cTREC 2001
#index507
#id507
#%
#! trec saw falling abeyance large web task strengthening broadening activities based million page wt corpus tasks topic relevance task like traditional trec ad hoc used queries taken real web search logs description narrative elds topic description inferred topic developers topics homepage nding task queries corresponded entity home

#* evaluating queries overweb accessible databases global page ordering
#@Amelie Marian
#t2002
#cProceedings of the 18th International Conference on Data Engineering
#index508
#id508
#%233,234,509,510,511,512,238,513,463,514,515,516,517,518,467,519
#! query web search engine usually consists list keywords search engine responds best pages query query model prevalent multimedia collections general plain relational data certain applications example consider relation information available restaurants including location price range diner overall food rating user queries relation simply specify users location target price range expect return best restaurants terms combination proximity user closeness match target price range overall food rating processing queries efficiently challenging number reasons critical reason web applications relation attributes available external web accessible form interfaces query repeatedly potentially large set candidate objects paper study process topk queries efficiently setting attributes users specify target values handled external autonomous sources variety access interfaces present algorithms processing queries evaluate thoroughly using synthetic real web accessible data

#* optimizing queries multimedia repositories
#@Surajit Chaudhuri,Luis Gravano
#t1996
#cSIGMOD Conference
#index509
#id509
#%
#! repositories multimedia objects having multiple types attributes image text increasingly common selection attributes typically produce just set objects traditional relational query model filtering grade match associated object indicating object matches selection condition ranking multimedia repositories allow access attributes object indexes investigate optimize processing queries multimedia repositories key issue choice indexes used search repository define execution space search minimal set indexes searched minimal general problem picking optimal plan search minimal execution space np hard solve problem efficiently predicates query independent problem optimizing queries ask ranked objects viewed cases evaluating selection conditions problems viewed extended filtering problem

#* evaluating selection queries
#@Surajit Chaudhuri,Luis Gravano
#t1999
#cVLDB
#index510
#id510
#%
#! applications users specify target values certain attributes requiring exact matches values returnfi instead result queries typically rank tuples best match given attribute valuesfi paper study advantages limitations processing query translating single range query traditional relational dbmss process cientlyfi particular study determine range query

#* probabilistic optimization queries
#@Donko Donjerkovic,Raghu Ramakrishnan
#t1999
#cVLDB
#index511
#id511
#%
#! problem finding best answers query quickly finding answers increasing importance relational databases applied multimedia decision support domains approach efficiently answering queries augment query additional selection prunes away unwanted portion answer set risk selection returns fewer desired number answers execution restarted

#* combining fuzzy information multiple systems
#@Ronald Fagin
#t1999
#cJ. Comput. Syst. Sci.
#index512
#id512
#%
#! traditional database result query set values values satisfy query data servers queries based image content text retrieval systems result query sorted list example case queries based image content query ask objects particular shade red result query sorted list objects database sorted

#* wsq dsq practical approach combined querying databases web
#@Roy Goldman,Jennifer Widom
#t2000
#cSIGMOD Conference
#index513
#id513
#%
#! present wsq dsq pronounced wisk disk new approach combining query facilitiesof traditional databases existing search engines webfi wsq web supported database queries leverages results web searches enhance sql queries relational database dsq database supported web queries uses information stored database enhance explainweb searchesfi paper focuses primarily wsq describing simple low overhead way

#* predicate migration optimizing queries expensive predicates
#@Joseph M. Hellerstein,Michael Stonebraker
#t1993
#cSIGMOD Conference
#index514
#id514
#%
#! traditional focus relational query optimization schemes choice join methods join orders restrictions typically handled query optimizers predicate pushdown rules apply restrictions random order joins possible rules work assumption restrictions essentially zero time operation today extensible object oriented database systems allow users define time consuming functions used query restriction join predicates furthermore sql long supported subquery predicates arbitrarily time consuming check restrictions considered zero time operations model query optimization enhanced paper develop theory moving expensive predicates query plan total cost plan including costs joins restrictions minimal present algorithm implement theory results implementation postgres experience newly enhanced postgres orders magnitude faster plans generated traditional query optimizer additional complexity considering expensive predicates optimization manageably small

#* prefer efficient execution multi parametric ranked queries
#@Vagelis Hristidis,Nick Koudas,Yannis Papakonstantinou
#t2001
#cSIGMOD Conference
#index515
#id515
#%
#! users need optimize selection objects ap propriately weighting importance multiple object tributes optimization problems appear opera tions research applied mathematics everydaylife buyer select home weighted functionof number attributes like distance office itsprice area capture queries definition preferencequeries use weight function relations attributesto derive score tuple database systems cannotefficiently produce results preference query cause need evaluate weight function tu ples relation prefer answers preference queriesefficiently using materialized views pre processed stored result preference query beproduced pipelined fashion using materialized view excellent performance deliveredgiven reasonable number materialized views pro vide algorithm selects number views precom pute materialize given space constraints implemented algorithms proposed pa prototype called prefer operateson commercial database management wepresent results performance comparison compar ing algorithms prior approaches using syntheticdatasets results indicate proposed algorithmsare superior performance compared approaches preprocessing preparation materialized views aswell execution time

#* optimizing disjunctive queries expensive predicates
#@Alfons Kemper,Guido Moerkotte,Klaus Peithner,Michael Steinbrunn
#t1994
#cSIGMOD Conference
#index516
#id516
#%
#! work propose assess technique called bypassprocessing optimizing evaluation disjunctivequeries expensive predicatesfi technique particularlyuseful optimizing selection predicates containterms evaluation costs vary tremendously efigfi evaluation nested subquery invocation auser defined function object oriented extended relationalmodel orders magnitude expensivethan attribute access

#* incremental join queries ranked inputs
#@Apostol Natsev,Yuan-Chi Chang,J.R. Smith,Chung-Sheng Li,Jeffrey Scott Vitter
#t2001
#cVLDB
#index517
#id517
#%
#! paper investigates problem incremental joins multiple ranked data sets join condition list arbitrary user defined predicates input tuples problem arises important applications dealing ordered inputs multiple ranked data sets requiring solutions use multimedia applications motivating examples problem equally applicable traditional database applications involving optimal resource allocation scheduling decision making ranking propose algorithmj enables querying ordered data sets imposing arbitrary userdefined join predicates basic version algorithm does use random access pa variation exploit available indexes efficient random access based join predicates special case includes join scenario considered fagin joins based identical keys case algorithms perform efficiently fagins main contribution generalization join scenarios previously unsupported including cases random access algorithm possible lack unique keys addition support multiple join levels nested join hierarchies norm modeling multimedia data approximation versions algorithms finally strong optimality results proposed algorithms study performance empirically

#* query processing issues image multimedia databases
#@Surya Nepal,M. V. Ramakrishna
#t1999
#cICDE
#index518
#id518
#%
#! multimedia databases attracted academic industrialinterest systems qbic content basedimage retrieval ibm releasedfi suchsystems essential effectively efficiently use existing large collections image data modern computingenvironmentfi aim systems enableretrieval images based contentsfi problemhas brought decades old database imageprocessing communitiesfias

#* numerical recipes art scientific computing
#@William H. Press
#t1992
#cbook
#index519
#id519
#%
#! book like predecessor edition supposed teach methods numerical computing practical efficient insofar possible elegant presume book reader particular tasks want view job educating proceed occasionally try reroute briefly particularly beautiful road large guide main highways lead practical destinations book fearlessly editorializing telling shouldnt prescriptive tone results conscious decision hope irritating claim advice infallible reacting tendency textbook literature computation discuss possible method invented offering practical judgment relative merit offer practical judgments gain experience form opinion reliable advice presume able read programs language version numerical recipes second edition book numerical recipes fortran second edition separately available prefer program language earlier editions numerical recipes pascal numerical recipes routines examples basic available containing additional material second edition versions fortran versions perfectly serviceable pascal basic language choice

#* exploiting hidden sentiment links reviews
#@Qi Su, Xinying Xu,  Honglei Guo, Zhili Guo, Xian Wu, Xiaoxun Zhang, Bin Swen, Zhong Su
#t1000
#cWWW
#index520
#id520
#%521,522,523,524,525,396,410,146,526,148,527,150,528,529,399,530,531,154,532
#! past years research effort devoted opinion mining product reviews recently research ort devoted feature level opinion mining instead review level identifying semantic relatedness product feature opinion words context methods usually utilize explicit relatedness product feature words opinion words reviews approach based explicit adjacency loss hidden sentiment association product features opinion words paper propose novel unsupervised mutual reinforcement approach deal feature level product opinion mining problem specially approach clusters product features opinion words simultaneously iteratively fusing content information sentiment link information framework based product feature categories opinion word groups construct sentiment association set groups data objects identifying strongest sentiment links knowledge multi source incorporated enhance clustering procedure based pre constructed association set approach largely predict opinions relating different product features case explicit appearance product feature words reviews provides accurate opinion evaluation experimental results demonstrate method outperforms state art algorithms

#* evaluating wordnetbased measures lexical semantic relatedness
#@Alexander Budanitsky,Graeme Hirst,Er Budanitsky Graeme Hirst
#t2004
#cComputational Linguistics
#index521
#id521
#%
#! quantification lexical semantic relatedness applications nlp different measures proposed evaluate measures use wordnet central resource comparing performance detecting correcting real word spelling errors information content based measure proposed jiang conrath superior proposed hirst st onge leacock chodorow lin resnik addition explain distributional similarity adequate proxy lexical semantic relatedness

#* noun phrase coreference clustering
#@Claire Cardie,Kiri Wagstaff
#t1999
#cProceedings of the 1999 SIGDAT Conference on Empirical Methods in Natural Language Processing and Very Large Corpora
#index522
#id522
#%
#! paper introduces new unsupervised algorithm noun phrase coreference resolution differs existing methods views corererence resolution clustering task evaluation muc coreference resolution corpus algorithm achieves measure placing firmly worst best systems muc evaluation importantly clustering approach outperforms muc treat coreference resolution learning problem clustering algorithm appears provide flexible mechanism coordinating application context independent context dependent constraints preferences accurate partitioning noun phrases coreference equivalence classes

#* mining peanut gallery opinion extraction semantic classication product reviews
#@K Dave,S Lawrence,DM Pennock
#t2003
#cInternational World Wide Web Conference
#index523
#id523
#%
#! web contains wealth product reviews sifting daunting task ideally opinion mining tool process set search results given item generating list product attributes quality features aggregating opinions poor mixed good begin identifying unique properties problem develop method automatically distinguishing positive negative reviews classifier draws information retrieval techniques feature extraction scoring results various metrics heuristics vary depending testing situation best methods work better traditional machine learning operating individual sentences collected web searches performance limited noise ambiguity context complete web based tool aided simple method grouping sentences attributes results qualitatively quite useful

#* summarizing visualizing arguments subjective documents supporting decision making
#@Atsushi Fujii,Tetsuya Ishikawa
#t2006
#cWorkshop On Sentiment And Subjectivity In Text
#index524
#id524
#%
#! world wide web volume subjective information opinions reviews increasing rapidly trends rules latent large set subjective descriptions potentially useful decision making purposes paper propose method summarizing subjective descriptions specifically opinions japanese visualize pro arguments target topic japan introduce summertime users summarize arguments topic order choose reasonable standpoint decision making evaluate called opinionreader experimentally

#* chinese named entity recognition based multilevel linguistic features
#@Honglei Guo,Jianmin Jiang,Gang Hu,Tong Zhang
#t2004
#cIJCNLP
#index525
#id525
#%
#! paper presents chinese named entity recognition employs robust risk minimization rrm classification method incorporates advantages character based word based models experiments large scale corpus significant performance enhancements obtained integrating various linguistic information chinese word segmentation semantic types speech named entity triggers basic chinese character based model novel feature weighting mechanism employed obtain useful cues important linguistic features overcome limitation computational resources building high quality named entity recognition large scale corpus informative samples selected active learning approach

#* methods classication analysis multivariate observations
#@James B.MacQueen
#t1967
#cProcedings of the Fifth Berkeley Symposium on Math, Statistics, and Probability
#index526
#id526
#%
#! paper describes number applications means procedure classifying random sample points sub procedure consists starting groups consist single random point adding points group mean point nearest point added group mean group adjusted account new point stage fact means group sample processed way points classified basis nearness final means portions result tend fficient sense having low class variance applications suggested problems non linear prediction efficient communication non parametric tests independence similarity grouping automatic file construction extension methods general metric spaces indicated

#* discovering word senses text
#@Patrick Pantel,Dekang Lin
#t2002
#cKDD
#index527
#id527
#%
#! inventories manually compiled dictionaries usually serve source word senses include rare senses missing corpus domain specific senses present clustering algorithm called cbc clustering committee automatically discovers word senses text initially discovers set tight clusters called committees scattered similarity space centroid members committee used feature vector cluster proceed assigning words similar clusters assigning element cluster remove overlapping features element allows cbc discover frequent senses word avoid discovering duplicate senses cluster word belongs represents senses present evaluation methodology automatically measuring precision recall discovered senses

#* cws comparative web search
#@Jian-Tao Sun,X Wang,Dou Shen,Hua-Jun Zeng,Zheng Chen
#t2006
#cWWW
#index528
#id528
#%
#! paper define study novel search problem comparative web search cws task cws seek relevant comparative information web help users conduct comparisons set topics called cws developed effectively facilitate web users comparison needs given set queries represent topics user wants compare characterized automatic retrieval ranking web pages incorporating relevance queries comparative contents contain automatic clustering comparative contents semantically meaningful themes extraction representative keyphrases summarize commonness differences comparative contents theme developed novel interface supports types view modes pair view displays result page level cluster view organizes comparative pages themes displays extracted phrases facilitate users comparison experiment results cws effective efficient

#* elements information theory
#@TM Cover,Joy A. Thomas
#t1991
#cbook
#index529
#id529
#%
#! essential topics information theory covered including entropy data compression channel capacity rate distortion network information theory hypothesis testing authors provide readers solid understanding underlying theory applications problem sets telegraphic summary end chapter assist readers historical notes follow chapter recap main points

#* constrained means clustering background knowledge
#@Kiri Wagstaff,Claire Cardie,Seth Rogers,Stefan Schrodl
#t2001
#cICML
#index530
#id530
#%
#! clustering traditionally viewed unsupervised method data analysis cases information problem domain available addition data instances paper demonstrate popular means clustering algorithm profitably modified make use information experiments artificial constraints data sets observe improvements clustering accuracy apply method real world problem automatically detecting road lanes gps data observe dramatic increases performance

#* reinforcing web object categorization interrelationships
#@Gui-Rong Xue,Yong Yu,Dou Shen,Qiang Yang,Hua-Jun Zeng,Zheng Chen
#t2006
#cData Min. Knowl. Discov.
#index531
#id531
#%
#! existing categorization algorithms deal homogeneous web objects consider interrelated objects additional features taking interrelationships types objects account focusing single aspect inter object relationship sufficient fully reveal true categories web objects paper propose novel categorization algorithm called iterative reinforcement categorization algorithm irc exploit interrelationship different types web objects web including web pages queries irc classifies interrelated web objects iteratively reinforcing individual classification results different types objects interrelationship experiments clickthrough log dataset msn search engine terms measure irc achieves improvement pure content based classification method achieves improvement query metadata based method improvement measure known virtual document based method experiments irc converges fast applicable real world applications

#* unied framework clustering heterogeneous web objects
#@Hua-Jun Zeng,Zheng Chen,Wei-Ying Ma
#t2002
#cWISE
#index532
#id532
#%
#! paper introduce novel framework clustering web data heterogeneous nature existing methods integrate heterogeneous data unified feature space flexibilities explore adjust contributing effect different heterogeneous information compromised contrast framework enables separate clustering homogeneous data entire process based respective features layered structure link information

#* fast robust fixed point algorithms independent component analysis
#@Aapo Hyvarinen
#t1999
#cIEEE Transactions on Neural Networks
#index533
#id533
#%534,535,536,537,529,538,539,540,541,542,543,544,545,546,547,548,549,550
#! independent component analysis ica statistical method transforming observed multidimen sional random vector components statistically independent possible paper use combination different approaches linear ica comons information theoretic approach projection pursuit approach using maximum entropy approximations differential en tropy introduce family new contrast objective functions ica contrast functions enable estimation decomposition minimizing mutual information estimation indi vidual independent components projection pursuit directions statistical properties estimators based contrast functions analyzed assumption linear mixture model shown choose contrast functions robust minimum variance finally intro duce simple xed point algorithms practical optimization contrast functions algorithms optimize contrast functions fast reliably

#* new learning algorithm blind source separation
#@S. Amari,A. Cichocki,H. H. Yang
#t1996
#cAdvances in Neural Information Processing Systems
#index534
#id534
#%
#! new line learning algorithm minimizes statistical dependency outputs derived blind separation mixed signals dependency measured average mutual information mi outputs source signals mixing matrix unknown number sources gram charlier expansion instead edgeworth expansion used evaluating mi natural gradient approach used minimize mi novel activation function

#* information maximization approach blind separation blind deconvolution
#@Anthony J. Bell,Terrence J. Sejnowski
#t1995
#cNeural Computation
#index535
#id535
#%
#! derive new self organizing learning algorithm maximizes information transferred network nonlinear units algorithm does assume knowledge input distributions defined zero noise limit conditions information maximization extra properties linear case linsker nonlinearities transfer function able pick higher order moments input distributions perform akin true redundancy reduction units output representation enables network separate statistically independent components inputs higher order generalization principal components analysis apply network source separation cocktail party problem successfully separating unknown mixtures speakers variant network architecture able perform blind deconvolution cancellation unknown echoes reverberation speech signal finally derive dependencies information transfer time delays suggest information maximization provides unifying framework problems blind signal processing

#* equivariant adaptive source separation
#@JF Cardoso,BH Laheld
#t1996
#cSignal Processing, IEEE Transactions on
#index536
#id536
#%
#! source separation consists recovering set independent signals mixtures unknown coefficients observed paper introduces class adaptive algorithms source separation implements adaptive version equivariant estimation henceforth called equivariant adaptive separation independence easi easi algorithms based idea serial updating specific form matrix updates systematically yields algorithms simple structure real complex mixtures importantly performance easi algorithm does depend mixing matrix particular convergence rates stability conditions interference rejection levels depend normalized distributions source signals closed form expressions quantities given asymptotic performance analysis theme equivariance stressed paper source separation problem underlying multiplicative structure parameter space forms matrix multiplicative group explore favorable consequences fact implementation performance optimization easi algorithms

#* independent component analysisa new concept signal processing
#@P. COMON
#t1994
#cSignal Processing archive
#index537
#id537
#%
#! applying unsupervised learning techniques like ica temporal decorrelation bss key question discovered projections reliable words error bars assess quality separation use resampling methods tackle questions experimentally proposed variance estimations strongly correlated separation error demonstrate reliability estimation used choose appropriate

#* projection pursuit algorithm exploratory data analysis
#@JH Friedman,J. W. Tukey
#t1974
#cComputers
#index538
#id538
#%
#! algorithm analysis multivariate data presented discussed terms specific examples algorithm seeks dimensional linear projections multivariate data relatively highly revealing

#* exploratory projection pursuit
#@JH Friedman
#t1987
#cJournal of the American statistical association
#index539
#id539
#%
#! exploratory projection pursuit epp tool exploratory data analysis theorem cram er wold allows concentrate bivariate projections data work diaconis freedman emphasizes search projections non normal density index functions used structure projection local maximum index function provides interesting dimensional views data variety index functions

#* experimental comparison neural ica algorithms
#@X Giannakopoulos,J Karhunen,E Oja
#t1998
#cProc. Int. Conf. on Artificial Neural Networks
#index540
#id540
#%
#! neural algorithms independent component analysis ica introduced lately computational properties systematically studied paper compare accuracy convergence speed computational load properties prominent neural semi neural ica algorithms comparison reveals interesting differences algorithms introduction independent component analysis ica unsupervised technique

#* robust statistics
#@PJ Huber,EM Ronchetti
#t1971
#cbook
#index541
#id541
#%
#! statistics classical methods rely heavily assumptions met practice particular assumed data normally distributed approximately central limit theorem relied produce normally distributed estimates unfortunately outliers data classical methods poor performance robust statistics seeks provide methods emulate classical methods unduly affected outliers small departures model assumptions

#* modern factor analysis
#@Harry H. Harman
#t1960
#cbook
#index542
#id542
#%
#! thoroughly revised edition harry harman authoritative text incorporates new advances science technology years author gives coverage theoretical applied aspects factor analysis foundations advanced techniques highly readable text welcomed researchers students working psychology statistics economics related disciplines

#* projection pursuit
#@PJ Huber
#t1985
#c The annals of Statistics
#index543
#id543
#%
#! article concerns automatic pp pp used todayficlearly non linear structures clusters separations unexpected shapes areinterestingfi general phrase non linear structures refers structures whichcannot detected easily using sample mean sample covariance matrixfi summary statistics like sample mean covariance reveal linear structures thelocation scale correlation structures data setfi

#* unit contrast functions independent component analysis statistical analysis
#@Aapo Hyvarinen
#t1997
#cNeural Networks for Signal Processing
#index544
#id544
#%
#! author introduced previously large family unit contrast functions used independent component analysis ica paper family analyzed mathematically case nite sample aspects estimators obtained using contrast functions considered asymptotic variance robustness outliers expression contrast function minimizes asymptotic variance obtained function probability densities

#* new approximations differential entropy independent component analysis projection pursuit
#@Aapo Hyvarinen
#t1997
#cNIPS
#index545
#id545
#%
#! derive order approximation density maximum entropy continuous random variable given number simple constraints results density expansion somewhat similar classical polynomial density expansions gram charlier edgeworth using approximation density approximation differential entropy derived approximation entropy exact robust outliers classical approximation

#* fast algorithm estimating overcomplete ica bases image windows
#@Aapo Hyvarinen,Razvan Cristescu,E Oja
#t1999
#cProc. Int. Joint Conf. on Neural Networks
#index546
#id546
#%
#! introduce fast method estimating overcomplete bases independent components image data based concept quasi orthogonality means high dimensional space large overcomplete set vectors orthogonal estimate overcomplete basis using oneunit ica algorithms forcing partial decorrelation different independent components method implemented using modification fastica algorithm leads computationally highly efficient method

#* independent component analysis general nonlinear hebbian like learning rules
#@Aapo Hyvarinen,E Oja
#t1998
#cSignal Processing
#index547
#id547
#%
#! number neural learning rules recently proposed paper fact ica performed simple hebbian anti hebbian learning rules weak relations information theoretical quantities suprisingly practically non linear function used learning rule provided sign hebbian anti hebbian term chosen correctly addition hebbian like mechanism weight vector

#* blind separation sources
#@P. COMON,C. JUTTEN,J. HERAULT
#t1991
#cSignal Processing
#index548
#id548
#%
#! blind source separation problem extract underlying source signals set linear mixtures mixing matrix unknown situation common acoustics radio medical signal image processing hyperspectral imaging areas suggest stage separation process priori selection possibly overcomplete signal dictionary instance wavelet frame learned dictionary sources assumed sparsely representable followed unmixing sources exploiting sparse representability consider general case sources mixtures derive efficient algorithm case nonovercomplete dictionary equal numbers sources mixtures experiments artificial signals musical sounds demonstrate significantly better separation known techniques

#* class neural networks independent component analysis
#@J Karhunen,E Oja,Liuyue Wang,Ricardo Vigario,Jyrki Joutsensalo
#t1995
#cIEEE transactions on neural networks
#index549
#id549
#%
#! independent component analysis ica recently developed useful extension standard principal component analysis pca ica model utilized mainly blind separation unknown source signals linear mixtures application source signals correspond coefficients ica expansion paper propose neural structures related multilayer feedforward networks performing complete ica basic ica network consists whitening separation basis vector estimation layers used blind source separation estimation basis vectors ica consider learning algorithms layer modify previous nonlinear pca type algorithms separation capabilities greatly improved proposed class networks yields good results test examples artificial real world data

#* extraction ocular artifacts eeg using independent component analysis
#@RN Vigrio
#t1997
#cElectroencephalography and clinical Neurophysiology
#index550
#id550
#%
#! laboratory information science helsinki university technology hut finland ricardo vigario hut fi eye activity main sources artefacts eeg meg recordings new approach correction disturbances presented using statistical technique independent component analysis technique separates components kurtosis amplitude distribution time distinguishing strictly periodical signals regularly occurring signals irregularly occurring signals category usually formed artefacts approach possible isolate pure eye activity eeg recordings including eog channels reduce brain activity subtracted measurements extracting portions eog signals

#* gap factor model discrete data
#@John F. Canny
#t2004
#cSIGIR
#index551
#id551
#%182,182,205,533,552,553,554,555,363,556,557
#! present probabilistic model document corpus combines desirable features previous models model called gap gamma poisson distributions random variable gap factor model gives approximate factorization document term matrix product matrices factors strictly non negative terms gap generative probabilistic model assigns finite probabilities documents corpus computed efficient simple em recurrence suitable choice parameters gap factorization maximizes independence factors used independent component algorithm adapted document data form gap model empirically analytically motivated gives accurate results probabilistic model measured perplexity retrieval model gap model projects documents terms low dimensional space themes models texts passages terms theme categories subject descriptors

#* independent component analysis algorithms applications
#@Aapo Hyvarinen,E Oja
#t2000
#cNeural Networks
#index552
#id552
#%
#! fundamental problem neural network research disciplines finding suitable representation multivariate data random vectors reasons computational conceptual simplicity representation sought linear transformation original data words component representation linear combination original variables known linear transformation methods include principal component analysis factor analysis projection pursuit independent component analysis ica recently developed method goal linear representation non gaussian data components statistically independent independent possible representation capture essential structure data applications including feature extraction signal separation paper present basic theory applications ica recent work subject

#* solution platos problem latent semantic analysis theory acquisition induction representation knowledge
#@TK Landauer,S. Dumais
#t1997
#cPsychological Review
#index553
#id553
#%
#! people know little information problem takes forms learning vocabulary text especially dramatic convenient case research new general theory acquired similarity knowledge representation latent semantic analysis lsa presented used successfully simulate learning psycholinguistic phenomena inducing global knowledge indirectly local occurrence data large body representative text lsa acquired knowledge vocabulary english comparable rate school children lsa uses prior linguistic perceptual similarity knowledge based solely general mathematical learning method achieves powerful inductive effects extracting right number dimensions represent objects contexts relations theories phenomena problems sketched

#* introduction latent semantic analysis
#@TK Landauer,PW Foltz,D Laham
#t1998
#cDiscourse processes
#index554
#id554
#%
#! latent semantic analysis lsa theory method extracting representing contextual usage meaning words statistical computations applied large corpus text landauer dumais underlying idea aggregate word contexts given word does does appear provides set mutual constraints largely determines similarity meaning words sets words adequacy lsas reflection human knowledge established variety ways example scores overlap humans standard vocabulary subject matter tests mimics human word sorting category judgments simulates wordcword passagecword lexical priming data reported following articles issue accurately estimates passage coherence learnability passages individual students quality quantity knowledge contained essay

#* learning human like knowledge singular value decomposition progress report
#@Thomas K. Landauer,D Laham,PW Foltz
#t1997
#cNIPS
#index555
#id555
#%
#! singular value decomposition svd viewed method unsupervised training network associates classes events reciprocally linear connections single hidden layer svd used learn represent relations large numbers words large numbers natural text passages occurred result dimensional semantic spaces trained newly added word passage represented vector similarities measured cosine contained angle vectors good accuracy simulating human judgments behaviors demonstrated performance multiple choice vocabulary domain knowledge tests emulation expert essay evaluations ways examples given kind knowledge extracted method applied

#* document clustering based non negative matrix factorization
#@Wei Xu,Xin Liu,Yihong Gong
#t2003
#cSIGIR
#index556
#id556
#%
#! paper propose novel document clustering method based non negative factorization term document matrix given document corpus latent semantic space derived non negative matrix factorization nmf axis captures base topic particular document cluster document represented additive combination base topics cluster membership document easily determined finding base topic axis document largest projection value experimental evaluations proposed document clustering method surpasses latent semantic indexing spectral clustering methods easy reliable derivation document clustering results document clustering accuracies

#* study smoothing methods language models applied ad hoc information retrieval
#@ChengXiang Zhai,J Lafferty
#t2001
#cSIGIR
#index557
#id557
#%
#! language modeling approaches information retrieval attractive promising connect problem retrieval language model estimation studied extensively application areas speech recognition basic idea approaches estimate language model document rank documents likelihood query according estimated language model core problem language model estimation smoothing adjusts maximum likelihood estimator correct inaccuracy data sparseness paper study problem language model smoothing influence retrieval performance examine sensitivity retrieval performance smoothing parameters compare popular smoothing methods different test collections

#* group topic discovery relations text
#@Xuerui Wang,Natasha Mohanty,A McCallum
#t2005
#cLinkKDD '05
#index558
#id558
#%559,560,561,182,562,563,564,565,566,567,568,569,570,571
#! present probabilistic generative model entity relationships textual attributes simultaneously discovers groups entities topics corresponding text block models relationship data studied social network analysis time simultaneously cluster modalities incorporating words associated certain relationships significantly joint inference allows discovery groups guided emerging topics vice versa present experimental results large data sets sixteen years bills senate comprising corresponding text voting records years similar data united nations comparison traditional separate latent variable models words blockstructures votes group topic models joint inference improves groups topics discovered categories subject descriptors

#* agglomerative clustering search engine query log
#@Doug Beeferman,Adam L. Berger
#t2000
#cKDD
#index559
#id559
#%
#! paper introduces technique mining collection ofuser transactions internet search engine discoverclusters similar queries similar urlsfi informationwe exploit clickthrough data record consistsof user query search engine urlwhich user selected candidates offeredby search engine viewing dataset bipartitegraph vertices corresponding queriesand

#* multi way distributional clustering pairwise interactions
#@Ron Bekkerman,Ran El-Yaniv,A McCallum
#t2005
#cICML
#index560
#id560
#%
#! present novel unsupervised learning scheme simultaneously clusters variables types documents words authors based pairwise interactions tween types observed occurrence data scheme multiple clustering systems generated aiming maximizing objective function measures multiple pairwise mutual information cluster variables implement idea pro pose algorithm interleaves clustering variables clustering variables local optimization correction routine focusing document clustering present extensive empirical study way way way applications scheme using real world datasets including news

#* deduplication group detection using links
#@Indrajit Bhattacharya,Lise Getoor
#t2004
#cACM SIGKDD Workshop on Link Analysis and Group Detection (LinkKDD)
#index561
#id561
#%
#! categories subject descriptors

#* information theoretic clustering
#@Inderjit S. Dhillon,Subramanyam Mallela,Dharmendra S. Modha
#t2003
#cKDD
#index562
#id562
#%
#! dimensional contingency occurrence tables arise frequently important applications text web log market basket data analysisfi basic problem contingency table analysis clustering simultaneous clustering rows columnsfi novel theoretical formulation views contingency table empirical joint probability distribution discrete random variables poses clustering problem optimization problem information theory

#* does europe make mind connections cliques compatibility countries eurovision song contest
#@Daniel Fenn,Omer Suleman,Janet Efstathiou,Neil F. Johnson
#t2005
#carXiv.org
#index563
#id563
#%
#! investigate complex relationships countries eurovision song contest recasting past voting data terms dynamical network despite british tendency feel distant europe analysis shows remarkably compatible tune european countries equally surprising finding core countries notably france significantly tune rest europe addition analysis enables confirm widely held belief unofficial cliques countries cliques expected ones existence explained solely grounds geographical proximity complexity emerges group self assessment process absence central controller speculate complexity representative real world situations groups agents establish inter relationships ultimately decide fate possible examples include groups individuals societies political groups governments

#* power parties cohesion competition european parliament
#@Hix Simon,Abdul Noury,Grard Roland
#t2005
#c2005	British Journal of Political Science
#index564
#id564
#%
#! cohesive political parties european parliament coalitions form answers questions central understanding impact european parliament european union policies questions central study legislative behaviour general collected total population roll votes european parliament elections end votes data growing party cohesion despite growing internal national ideological diversity european party groups distance parties left right dimension strongest predictor coalition patterns weconclude increased power european parliament meant increased power transnational parties increased internal party cohesion inter party competition

#* analyzing senate similarities networks clusters blocs
#@A Jakulin,W Buntine,Timothy La Pira,Holly Brasher
#t2003
#cPolitical Analysis
#index565
#id565
#%
#! analyze roll calls senate year employed methods used science community analysis genes surveys text information theoretic measures assess association pairs senators based votes cast furthermore evaluate influence voter postulating shannon information channel outcome voter matrix associations summarized using hierarchical clustering multi dimensional scaling link analysis discrete latent variable model identify blocs cohesive voters senate contrast continuous ideal point methods bloc voting model senate interpreted weighted vote able estimate empirical voting power individual blocs analysis

#* discovering latent classes relational data
#@Kemp Charles,Thomas L. Griffiths, Joshua B. Tenenbaum
#t2004
#cnull
#index566
#id566
#%
#! present framework learning relational knowledge aimof explaining people acquire intuitive theories physical biological socialsystems approach based generative relational model latent classes simultaneously determines kinds entities exist domain num ber latent classes relations classes possible likely model goes previous psychological models category learning whichconsider attributes associated individual categories relationships betweencategories apply domain general framework specific problems learn ing structure kinship systems learning causal theories

#* pcans model structure organization
#@KM Carley,D Krackhardt
#t1998
#cProceedings of the 1998 International Symposiumon Command and Control Research and Technology
#index567
#id567
#%
#! present network based approach characterize architectures terms domain elements individuals tasks resources characterizing possible relations elements results relational primitives precedence commitment resources assignment individuals tasks networks relations personnel skills linkingindividuals resources demonstrate utility model recharacterizingclassical organizational theory generating series testable hypotheses performance

#* stochastic link group detection
#@Jeremy Kubica,Andrew W. Moore,Jeff G. Schneider,Yiming Yang
#t2002
#cAAAI/IAAI
#index568
#id568
#%
#! link detection analysis long important social sciences government intelligence communityfi significant effort focused structural functional analysis known networksfi similarly detection individual links important usually techniques result known linksfi recently internet sources led flood circumstantial data provide probabilistic evidence linksfi occurrence news

#* topic role discovery social networks
#@A McCallum,A Corrada-Emmanuel,Xuerui Wang
#t2005
#cIJCAI
#index569
#id569
#%
#! previous work social network analysis sna modeled existence links en tity language content topics links present author recipient topic art model social network analysis learns topic distributions based direction sensitive messages sent en tities model builds latent dirichlet al location lda author topic model adding key attribute distribution ics conditioned distinctly sender recipient steering discovery topics accord ing relationships people

#* estimation prediction stochastic blockstructures
#@Krzysztof Nowicki,Tom A.B. Snijders
#t1996
#cJournal of the American Statistical Association
#index570
#id570
#%
#! paper extend approach snijders andnowicki case relations directedand arbirary set possible values number classes arbitraryfi vectorof attributes specifying class structure consideredto unobserved latent fi conditional thevector model using generalization pairdependentstochastic blockmodelfi model regardedas mixture model classes whichthe vertices

#* parliamentary group individual voting behavior finnish parliamentin year group cohesion voting similarity analysis
#@A Pajala,A Jakulin,W Buntine
#t1000
#cEMPTY
#index571
#id571
#%
#! group cohesion studies common analyses finnish parliament eduskunta published article provides firstly fresh group cohesion analysis using agreement index modified version classic rice index secondly advanced voting similarity analyses new easy understand way illustrating results provided agreement index operates parliamentary party group level voting similarity analyses able analyse illustrate individual mp level article partly methodological testing voting similarity methods provides insight recent voting behaviour eduskunta

#* hidden markov models molecular biology new algorithms applications
#@Pierre Baldi,Yves Chauvin,Tim Hunkapiller,Marcella A. McClure
#t1992
#cNIPS
#index572
#id572
#%
#! hidden markov models hmms applied important problems molecular biology introduce new convergent learning algorithm hmms unlike classical baum welch algorithm smooth applied line batch mode usual viterbi likely path approximation left right hmms insertion deletion states trained represent protein families including immunoglobulins kinases cases models

#* statistical decision theory bayesian analysis
#@J. O. Berger
#t1985
#cbook
#index573
#id573
#%
#! outstanding strengths book topic coverage references exposition examples problem sets book excellent addition mathematical statistician library bulletin american mathematical society new edition author added substantial material bayesian analysis including lengthy new sections important topics empirical hierarchical bayes analysis bayesian calculation bayesian communication group decision making changes book used self contained introduction bayesian analysis addition decision theoretic portion text updated including new sections covering modern topics minimax multivariate stein estimation

#* protein modeling using hidden markov models analysis globins
#@D Haussler,A Krogh,IS Mian,K Sjolander
#t1993
#cSystem Sciences
#index574
#id574
#%
#! apply hidden markov models hmms theproblem statistical modeling multiple alignmentof protein familiesfi variant expectation maximization em algorithm known viterbi algorithmis used obtain statistical model theunaligned sequencesfi detailed series experiments taken unaligned globin sequences produced statistical model entirely automaticallyfrom primary unaligned sequences using priorknowledge globin

#* study grammatical inference
#@James Jay Horning
#t1969
#cnull
#index575
#id575
#%
#! grammatical inference inductive process discovering acceptable grammar language basis finite samples language study goals devising useful inference procedures demonstrating sound formal basis procedures states general grammatical inference problem formal languages reviews previous work establishes definitions notation states position evaluation measures indicates solution particular class grammatical inference problems based assumed probabilistic structure

#* estimation stochastic context free grammars using inside outside algorithm
#@K. Lari,S. Young
#t1990
#cComputer Speech & Language
#index576
#id576
#%
#! combination transformation algorithm stochastic context free grammars inside outside algorithm allows define method estimation rule probabilities stochastic context free grammars time complexity inside outside algorithm transformation algorithm relates stochastic context free grammars characteristic grammar proper does single rules stochastic context free grammars chomsky normal form

#* learning automata ordered examples
#@Sara Porat,Jerome A. Feldman
#t1991
#cMachine Learning
#index577
#id577
#%
#! connectionist learning models considerable empirical success hard characterize exactly learn learning finite state languages fsl example strings domain extensively studied provide opportunity help understand connectionist learning major problem traditional fsl learning assumes storage examples violates connectionist principles paper presents provably correct algorithm inferring minimum state deterministic finite state automata fsa complete ordered sample using limited total storage storing example strings algorithm iterative strategy uses stage current encoding data considered far single sample string crucial advantages algorithm total space used course learning encoding finite prefix sample polynomial size inferred minimum state deterministic fsa algorithm relatively efficient time implemented importantly connectionist version algorithm preserves properties connectionist version requires structure usual models implemented using rochester connectionist simulator machine finite working storage iteratively identify fsl arbitrary presentations

#* introduction hidden markov models
#@Lawrence R. Rabiner, Bing-Hwang Juang
#t1986
#cASSP Magazine,IEEE
#index578
#id578
#%
#! basic theory markov chains known mathematicians engineers close years past decade applied explicitly problems speech processing major reasons speech models based markov chains developed recently lack method optimizing parameters markov model match observed signal patterns method proposed late immediately applied speech processing research institutions continued refinements theory implementation markov modelling techniques greatly enhanced method leading wide range applications models purpose tutorial paper introduction theory markov models illustrate applied problems speech recognition

#* universal prior integers estimation minimum description length
#@Jorma Rissanen
#t1983
#c Ann. Statist
#index579
#id579
#%
#! earlier introduced estimation principle calls minimization number bits required write observed data reformulated extend classical maximum likelihood principle principle permits estimation number parameters statistical models addition values way parameters appear models model structures principle rests new way interpret construct universal prior distribution integers makes sense parameter individual object truncated real valued parameters converted integers dividing precision prior determined universal prior integers optimizing precision

#* estimation inference compact coding
#@C. S. Wallace,P. R. Freeman
#t1987
#cJournal of the Royal Statistical Society. Series B (Methodological)
#index580
#id580
#%
#! systematic variation set data represented usual statistical model used encode data compact form possible considered purely random encoded form parts states inferred estimates unknown parameters model second states data using optimal code based data probability distribution implied parameter estimates choosing model estimates compact coding leads interesting general inference procedure strict form great generality nice properties computationally infeasible approximate form developed relation methods explored

#* motor imagery direct brain communication
#@G. Pfurtscheller, C. Neuper
#t2001
#cProceedings of the IEEE
#index581
#id581
#%
#! motor imagery modify neuronal activity primary sensorimotor areas similar way observable real executed movement eeg based brain interfaces bci based recording classification circumscribed transient eeg changes different types motor imagery imagination left hand right hand foot movement features band power adaptive autoregressive parameters extracted bipolar eeg recordings overlaying sensorimotor areas array electrodes located central neighboring areas classification features linear discrimination analysis neural networks used characteristic graz bci classifier set learning session updated sessions online feedback using procedure rapid prototyping result discrimination brain states leftversus right hand movement imagination reached days training time tetraplegic patient able operate eeg based control hand orthosis nearly classification accuracy mental imagination specific motor commands

#* using timedependent neural networks eeg classification
#@Ernst Haselsteiner,G. Pfurtscheller
#t2000
#cIEEE transactions on rehabilitation engineering
#index582
#id582
#%
#! paper compares different topologies neural networks used classify single trial electroencephalograph eeg data brainccomputer interface bci short introduction time series classification given used classifiers described standard multilayer perceptrons mlps used standard method classification compared finite impulse response fir mlps use fir filters instead static weights allow temporal processing inside classifier theoretical comparison architectures presented results bci experiment different subjects given discussed results demonstrate higher performance fir mlp compared standard mlp

#* eeg pattern recognitionarousal states detection classification
#@R. S. Huang,C. J. Kuo,L.-L. Tsai,O. T. C. Chen
#t1996
#cProc.IEEE Conf. Neural Netw 1996
#index583
#id583
#%
#! use electroencephalogram eeg detect arousal states humans eeg patterns fluctuating waking sleep described features extracted moving time windows significant feature mean frequency mf used arousal states detection fluctuations eeg mean frequency characterized hidden markov models hmms estimate possible arousal state hmm single current state transition probability considered global states transitions local contextual effects lce global contextual effects gce autoregressive hmms used estimate detect transitions waking sleep validity proposed model verified behavior measure correct rate subject responses auditory stimuli study estimated values mean frequency gce hmm high correlation behavior measure high recognition rate makes arousal states detection practicable furthermore lce hmms applicable artifacts rejection real time arousal states detection alarms given subject vigilant states turn drowsy states applications example used maintain long term vehicle driver arousal

#* hmm used offline classification eeg data
#@B. Obermaier,Ch. Guger,G. Pfurtscheller
#t1999
#cBiomedizinsche Technik
#index584
#id584
#%
#! hidden markov models hmm introduced offline classification single trail eeg data brain interface bci hmms used classify hjorth parameters calculated bipolar eeg data recorded imagination left right hand movement effects different types hmms recognition rate discussed furthermore comparison results achieved linear discriminant ld hmm presented

#* gaussian observation hidden markov models eeg analysis
#@William D. Penny,Stephen J. Roberts
#t1998
#cTechnical Report
#index585
#id585
#%
#! using number synthetic data sets hmms detect changesin dc levels correlation frequency coherence typical nonstationarychanges eeg signalfi number hidden states hmmcan chosen using cluster analysis derived featuresfi experiments based onthe use hmms gaussian observation densities trained autoregressive ar multivariate autoregressive mar coefficientsfi extraction

#* coupled hidden markov models modeling interactive processes
#@M. Brand,N. Oliver, A. Pentland 1997
#t0
#cnull
#index586
#id586
#%
#! present methods coupling hidden markov models hmms model systems multiple interacting processes resulting models multiple state variables temporally coupled matrices conditional probabilities introduce deterministic cn approximation maximum posterior map state estimation enables fast classification parameter estimation expectation maximization heads dynamic programming algorithm samples highest

#* estimation coupled hidden markov models application biosignal interaction modelling
#@Iead Rezek,Stephen J. Roberts
#t2000
#c2000	10th IEEE Workshop on Neural Networks for Signal Processing (NNSP 2000)
#index587
#id587
#%
#! coupled hidden markov models chmm new tool model interactions state space observation space reveal coupling classical tools correlation fail paper derive maximum likelihood equations chmm parameters using expectation maximisation algorithm use models demonstrated simulated data biomedical signal analysis

#* input output hmms sequence processing
#@Yoshua Bengio,Paolo Frasconi
#t1995
#cIEEE Transactions on Neural Networks
#index588
#id588
#%
#! consider problems sequence processing propose solution based discrete state model order represent past context introduce recurrent connectionist architecture having modular structure associates subnetwork state model statistical interpretation input output hidden markov model iohmm trained estimation maximization em generalized em gem algorithms considering state trajectories missing data decouples temporal credit assignment actual parameter estimation model presents similarities hidden markov models hmms allows map input sequences output sequences using processing style recurrent neural networks iohmms trained using discriminant learning paradigm hmms potentially taking advantage em algorithm demonstrate iohmms suited solving grammatical inference problems benchmark problem experimental results presented seven tomita grammars showing adaptive models attain excellent generalization

#* factorial hidden markov models
#@Zoubin Ghahramani,Michael I. Jordan
#t1997
#cMachine Learning
#index589
#id589
#%
#! present framework learning hidden markov models distributed staterepresentations framework derive learning algorithm based theexpectation maximization em procedure maximum likelihood estimation anal ogous standard baum welch update rules step algorithm exactand solved analytically combinatorial nature hiddenstate representation exact step intractable simple tractable mean eldapproximation derived empirical results set problems suggest mean eld approximation gibbs sampling viable alternatives computa tionally expensive exact algorithm

#* mixed memory markov models decomposing complex stochastic processes mixtures simpler ones
#@Lawrence K. Saul,Michael I. Jordan
#t1999
#cMachine Learning
#index590
#id590
#%
#! fi study markov models state spaces arise cartesian product morediscrete random variablesfi parameterize transition matrices models convexcombination mixture simpler dynamical modelsfi parameters models admit simpleprobabilistic interpretation fitted iteratively expectation maximization em procedurefi wederive set generalized baum welch updates factorial hidden markov models

#* coupled hidden markov models complex action recognition
#@M. Brand,N. Oliver,Alex Pentland
#t1997
#cCVPR
#index591
#id591
#%
#! present algorithms coupling training hidden markov models hmms model interacting processes demonstrate superiority conventional hmms vision task classifying handed actions hmms successful framework perceptual computing modeling classifying dynamic behaviors popular offer dynamic time warping training algorithm clear bayesian semantics markovian framework makes strong restrictive assumptions generating signal single process having small number states extremely limited state memory single process model inappropriate vision speech applications resulting low ceilings model performance coupled hmms provide efficient way resolve problems offer superior training speeds model likelihoods robustness initial conditions

#* modeling freeway traffic coupled hmms
#@Jaimyoung Kwon,Kevin Murphy
#t2000
#cTech. Rep
#index592
#id592
#%
#! consider problem modeling loop detector data collected fromfreewaysfi data vector time series contains speed ofvehicles averaged second sampling window number ofsites freewayfi assume measured speed locationis generated hidden discrete variable represents underlyingstate efigfi congested free flowing traffic point inspace timefi assume hidden variables

#* event coupled hidden markov models
#@Trausti T. Kristjansson,Brendan J. Frey,T.Huang
#t2000
#cIEEE International Conference on Multimedia and Expo (I)
#index593
#id593
#%
#! inferences time series data greatly enhancedby taking account multiple modalitiesfi cases audio speech corresponding video lipgestures different time series tightly coupledfi weare interested loosely coupled time series theonset events coupled timefi present extensionof forward backward algorithm used inferenceand learning event coupled hidden markov modelsand results

#* approximate learning dynamic models
#@Xavier Boyen,Daphne Koller
#t1998
#cNIPS
#index594
#id594
#%
#! inference key component learning probabilistic models partiallyobservable data learning temporal models manyinference phases requires complete traversal potentially verylong sequence furthermore data structures propagated procedurecan extremely large making process demandingfiin approximate inference algorithm monitoringstochastic processes prove bounds approximation errorfi

#* factored frontier algorithm approximate inference dbns
#@Kevin P. Murphy,Yair Weiss
#t2001
#cUAI
#index595
#id595
#%
#! factored frontier ff algorithm simple approximate inference algorithm dynamic bayesian networks dbns similar fully factorized version boyen koller bk algorithm instead doing exact update step followed marginalisation projection works factored distributions applied models exact update step intractable ff equivalent iteration loopy belief

#* growth transformations functions manifolds
#@LE Baum,GR Sell
#t1968
#cPacific J. Math
#index596
#id596
#%
#! paper look problem maximizing function defined manifold shall primarily concerned case certain polyhedron euclidean space rn polynomial nonnegative coefficients defined rn results valid greater generality

#* maximum likelihood estimation multivariate mixture observations markov chains
#@Bing-Hwang Juang,Stephen E. Levinson,M. Mohan Sondhi
#t1986
#cIEEE Transactions on Information Theory
#index597
#id597
#%
#! use probabilistic functions markov chain model certain parameterizations speech signal extend estimation technique liporace eases multivariate mixtures gaussian sums products mixtures problems relate liporace original framework

#* identifying comparative sentences text documents
#@Nitin Jindal,Bing Liu
#t2006
#cSIGIR
#index598
#id598
#%599,600,601,404,602,405,603,408,409,396,141,207,604,146,605,606,149,607,150,151,154,608
#! paper studies problem identifying comparative sentences text documents problem related quite different sentiment opinion sentence identification classification sentiment classification studies problem classifying document sentence based subjective opinion author important application area sentiment opinion identification business intelligence product manufacturer wants know consumers opinions products comparisons hand subjective objective furthermore comparison concerned object isolation instead compares object example opinion sentence sound quality cd player poor example comparative sentence sound quality cd player good cd player clearly sentences different information language constructs quite different identifying comparative sentences useful practice direct comparisons convincing ways evaluation important opinions individual object paper proposes study comparative sentence identification problem categorizes comparative sentences different types presents novel integrated pattern discovery supervised learning approach identifying comparative sentences text documents experiment results using types documents news articles consumer reviews products internet forum postings precision recall detailed results given paper

#* mining sequential patterns
#@Rakesh Agrawal,Ramakrishnan Srikant
#t1995
#cICDE
#index599
#id599
#%
#! given large database customer transactions transaction consists customer id transaction time items bought transaction introduce problem mining sequential patterns databases present algorithms solve problem empirically evaluate performance using synthetic data proposed algorithms apriorisome aprioriall comparable performance albeit apriorisome performs little better

#* sequential pattern mining using bitmap representation
#@Jay Ayres,Jason Flannick,Johannes Gehrke,Tomi Yiu
#t2002
#cKDD
#index600
#id600
#%
#! introduce new algorithm mining sequential patterns algorithm especially efficient sequential patterns database long introduce novel depth search strategy integrates depth traversal search space effective pruning mechanisms implementation search strategy combines vertical bitmap representation database efficient support counting salient feature algorithm incrementally outputs new frequent itemsets online fashion thorough experimental evaluation algorithm standard benchmark data literature algorithm outperforms previous work order magnitude

#* simple rule based speech tagger
#@Eric Brill
#t1992
#cANLP
#index601
#id601
#%
#! automatic speech tagging area natural language processing statistical techniques successful rule based methods paper present sim ple rule based speech tagger automatically acquires rules tags accuracy coinparable stochastic taggers rule based tagger advantages taggers including vast reduction stored information required perspicuity sinall set meaningful rules

#* xtag wide coverage grammar english
#@Christy Doran,Dania Egedi,Beth Ann Hockey,B. Srinivas,Martin Zaidel
#t1994
#cProceedings of the 17 th International Conference on Computational Linguistics (COLING '94)
#index602
#id602
#%
#! paper presents xtag grammar development tool based tree adjoining grammar tag formalism includes wide coverage syntactic grammar english various components discussed preliminary evaluation results parsing various corpora given results comparison xtag ibm statistical parser alvey natural language tool parser given introduction xtag large going project develop

#* extracting knowledge evaluative text
#@Giuseppe Carenini,Raymond T. Ng,Ed Zwart
#t2005
#cK-CAP
#index603
#id603
#%
#! capturing knowledge free form evaluative texts entity challenging task new techniques feature extraction polarity determination strength evaluation proposed feature extraction particularly important task provides underpinnings extracted knowledge work paper introduces improved method feature extraction draws existing unsupervised method including user specific prior knowledge evaluated entity turn task feature extraction term similarity mapping crude learned features user defined taxonomy entity features results promise terms accuracy mapping reduction semantic redundancy crude features

#* summarization tracking news blog corpora
#@
#t1000
#cEMPTY
#index604
#id604
#%
#! humans like express opinions eager know opinions automatically mining organizing opinions heterogeneous information sources useful individuals organizations governments opinion extraction opinion summarization opinion tracking important techniques understanding opinions opinion extraction mines opinions word sentence document levels articles opinion summarization summarizes opinions articles telling sentiment polarities degree correlated events paper news web blog articles investigated trec ntcir articles collected web blogs serve information sources opinion extraction documents related issue animal cloning selected experimental materials algorithms opinion extraction word sentence document level proposed issue relevant sentence selection discussed topical opinionated information summarized opinion summarizations visualized representative sentences text based summaries different languages different sources compared finally opinionated curve showing supportive nonsupportive degree timeline illustrated opinion tracking

#* web data mining exploring hyperlinks contents usage data
#@Olfa Nasraoui
#t2008
#cSIGKDD Explorations
#index605
#id605
#%
#! association rule mining important model data mining mining algorithms discover item associations rules data satisfy user specified minimum support minsup minimum confidence minconf constraints minsup controls minimum number data cases rule cover minconf controls predictive strength rule minsup used database model implicitly assumes items data nature similar frequencies data seldom case reallife applications applications items appear frequently data rarely appear minsup set high rules involve rare items rules involve frequent rare items minsup set low cause combinatorial explosion frequent items associated possible ways dilemma called rare item problem paper proposes novel technique solve problem technique allows user specify multiple minimum supports reflect natures items varied frequencies database rule mining different rules need satisfy different minimum supports depending items rules experiment results technique effective

#* sentiment analysis capturing favorability using natural language processing
#@Tetsuya Nasukawa,Jeonghee Yi
#t2003
#cK-CAP
#index606
#id606
#%
#! paper illustrates sentiment analysis approach extract sentiments associated polarities positive negative specific subjects document instead classifying document positive negative essential issues sentiment analysis identify sentiments expressed texts expressions indicate positive favorable negative unfavorable opinions subject order improve accuracy sentiment analysis important properly identify semantic relationships sentiment expressions subject applying semantic analysis syntactic parser sentiment lexicon prototype achieved high precision depending data finding sentiments web pages news articles

#* mining sequential patterns pattern growth prefixspan approach
#@Jian Pei,Jiawei Han,Behzad Mortazavi-Asl,Jianyong Wang,Helen Pinto,Qiming Chen,Umeshwar Dayal,Meichun Hsu
#t2004
#cIEEE Trans. Knowl. Data Eng.
#index607
#id607
#%
#! sequential pattern mining important data mining problem broad applications difficult problem mining generate examine combinatorially explosive number intermediate subsequences previously developed sequential pattern mining methods gsp explore candidate generation test approach reduce number candidates examined approach efficient mining large sequence databases having numerous patterns long patterns paper propose projection based sequential pattern growth approach efficient mining sequential patterns approach sequence database recursively projected set smaller projected databases sequential patterns grown projected database exploring locally frequent fragments based initial study pattern growth based sequential pattern mining freespan propose efficient method called psp offers ordered growth reduced projected databases improve performance pseudoprojection technique developed prefixspan comprehensive performance study shows prefixspan cases outperforms priori based algorithm gsp freespan spade sequential pattern mining algorithm adopts vertical data format prefixspan integrated pseudoprojection fastest tested algorithms furthermore mining methodology extended mining sequential patterns user specified constraints high promise pattern growth approach lead extension efficient mining kinds frequent patterns frequent substructures

#* cross collection mixture model comparative text mining
#@ChengXiang Zhai,Atulya Velivelli,Bei Yu
#t2004
#cKDD
#index608
#id608
#%
#! problem refer comparative text mining ctm given set comparable text collections task comparative text mining discover latent common themes collections summarize similarity di erences collections common theme general problem subsumes interesting applications including business intelligence opinion summarization propose generative probabilistic mixture model comparative text

#* incorporate syntactic knowledge opinion mining
#@Guang Qiu
#t2008
#cconjunction with WWW'08
#index609
#id609
#%610,145,143,224,611,396,612,613,614,615,154,149,616,617,618,619
#! development accessibly internet content web changed user generated content ugc kind novel media content produced end users taken past years revolution web flourish especially impressive china adoption ugc proven beneficial numbers traditional tasks dramatic increase volume data prevents users utilizing manual way automatic mining approaches demanded opinion mining recent data mining technique crossroad information retrieval computational linguistics pretty suitable kind information processing paper dedicate work main subtasks opinion mining topic extraction sentiment classification propose approaches issues respectively chinese based consideration syntactic knowledge blog data typical application ugc evaluating data experiments results approaches tasks promising introduction future plans stemmed work paper intelligent advertisement placement ugc

#* rise user generated content
#@Frank Bell
#t2007
#cnull
#index610
#id610
#%
#! web dominated consumer end user browser internet connection publish content power speak reviewing product service developing brands capital required

#* opinion holder extraction author authority viewpoints
#@Yohei Seki
#t2007
#cSIGIR
#index611
#id611
#%
#! opinion holder extraction research important discriminating opinions viewed different perspectives paper experience participation ntcir opinion analysis pilot task focusing opinion holder extraction results japanese english approach opinion holder extraction based discrimination author authority viewpoints opinionated sentences evaluation results fair respect japanese documents

#* topic sentiment mixture modeling facets opinions weblogs
#@Qiaozhu Mei,Xu Ling,Matthew Wondra,Hang Su,ChengXiang Zhai
#t2007
#cWWW
#index612
#id612
#%
#! paper define problem topic sentiment analysis weblogs propose novel probabilistic model capture mixture topics sentiments simultaneously proposed topic sentiment mixture tsm model reveal latent topical facets weblog collection subtopics results ad hoc query associated sentiments provide general sentiment models applicable ad hoc topics specifically designed hmm structure sentiment models topic models estimated tsm utilized extract topic life cycles sentiment dynamics empirical experiments different weblog datasets approach effective modeling topic facets sentiments extracting dynamics weblog collections tsm model quite general applied text collections mixture topics sentiments potential applications search result summarization opinion tracking user behavior prediction

#* learning subjective adjective corpora
#@Janyce Wiebe
#t2000
#cAAAI
#index613
#id613
#%
#! subjectivity tagging distinguishing sentences used present opinions evaluations sentences used objectively present factual information numerous applications subjectivity tagging relevant including information extraction information retrieval paper identifies strong clues subjectivity using results method clustering words according distributional similarity lin seeded small detailed manual annotation

#* automatic retrieval clustering similar words
#@Dekang Lin
#t1998
#cCOLING-ACL
#index614
#id614
#%
#! bootstrapping semantics text greatest challenges natural language learning earlier research showed possible automatically identify words semantically similar given word based syntactic collocation patterns words present approach goes step obtaining tree structure similar words different senses given word identified different subtrees submission type paper topic

#* measuring praise criticism inference semantic orientation association
#@Peter D. Turney,Michael L. Littman
#t2003
#cCoRR
#index615
#id615
#%
#! evaluative character word called semantic orientation positive semantic orientation indicates praise honest intrepid negative semantic orientation indicates criticism disturbing superfluous semantic orientation varies direction positive negative degree mild strong automated measuring semantic orientation application text classification text filtering tracking opinions online discussions analysis survey responses automated chat systems chatbots paper introduces method inferring semantic orientation word statistical association set positive negative paradigm words instances approach evaluated based different statistical measures word association pointwise mutual information pmi latent semantic analysis lsa method experimentally tested words including adjectives adverbs nouns verbs manually labeled positive words negative words method attains accuracy test set accuracy rises algorithm allowed abstain classifying mild words

#* utility linguistic rules opinion mining
#@Xiaowen Ding,Bing Liu
#t2007
#cSIGIR
#index616
#id616
#%
#! online product reviews important opinion sources web paper studies problem determining semantic orientations positive negative opinions expressed product features reviews existing approaches use set opinion words purpose semantic orientations words context dependent paper propose use linguistic rules deal problem new opinion aggregation function extensive experiments rules function highly effective called opinion observer built

#* extracting opinion topics chinese opinions using dependence grammar
#@
#t1000
#cEMPTY
#index617
#id617
#%
#! previous work opinion sentiment mining focuses sentiment classification postulation topics identified prior assumption fails reality advertising topics users commenting crucial corresponding advertisements promoted advertisers idea users referring paper propose rule based approach extracting topics opinion sentences given sentences identified texts advance build sentiment dictionary define rules based syntactic roles words using dependence grammar considered suitable chinese natural language parsing experiments encouraging results

#* efficient syntactic tagging toll corpora
#@Ming Zhou Changning Huang
#t1000
#cEMPTY
#index618
#id618
#%
#! tree bank important resoures mt linguistics researches requires large number sentences annotated syntactic information time consuming troublesome difficult consistency annotation manually paper presented new technique semi automatic tagging chinese text takes input chinese text outputs syntactically tagged sentence dependency tree use dependency grammar employ stack based shift reduce context dependent parser tagging mechanism works human machine cooperative way machine acquire tagging rules human intervention automation level improved step step accumulating rules annotation addition good consistency tagging guaranteed

#* parsing english link grammar
#@Daniel Dominic Sleator,David Temperley
#t1995
#cCoRR
#index619
#id619
#%
#! develop formal grammatical called link grammar english grammar encoded algorithms efficiently parsing link grammar expressive power link grammars equivalent context free grammars encoding natural language grammars appears easier new written program general link parsing written link grammar english language performance

#* novel word clustering algorithm based latent semantic analysis
#@Bellegarda,J.R.Butzberger,Janyce Wiebe
#t1996
#cICASSP
#index620
#id620
#%
#! new approach proposed clustering words given vocabulary method based paradigm formulated context information retrieval called latent semantic analysis paradigm leads parsimonious vector representation word suitable vector space familiar clustering techniques applied distance measure selected space arises naturally problem formulation preliminary experiments indicate clusters produced intuitively satisfactory clusters semantic nature approach prove useful complement conventional class based statistical language modeling techniques

#* improving retrieval information external sources
#@S. Dumais
#t1991
#cBehavior Research Methods, Instruments and Computers
#index621
#id621
#%
#! major barrier successful retrieval external sources electronic databases tremendous variability words people use objects fact different authors use different words essentially idea means relevant objects missed conversely fact word canbe used refer different things means irrelevant objects retrieved statistical method called latent semantic indexing models implicit higher order structure association words objects improves retrieval performance additional large performance improvements achieved use differential term weighting iterative retrieval methods

#* document space models using latent semantic analysis
#@Yoshihiko Gotoh,Steve Renals
#t1997
#c Eurospeech
#index622
#id622
#%
#! paper approach constructing mixture language models lms based notion semantics discussed end technique known latent semantic analysis lsa used approach encapsulates corpus derived semantic information able model varying style text using information corpus texts clustered unsupervised manner mixture lms automatically created work builds previous work field information

#* improving predicting performance statistical language models sparse domains
#@Rukmini Iyer,Dr. Mari Ostendorf
#t1998
#cnull
#index623
#id623
#%
#! standard statistical language models gram models represent probability word sequences suffer sparse data problems tasks large amounts domain specific text available thesis focuses improving estimation domain dependent gram models using domain text data previous approaches estimating language models multi domain data accounted characteristic variations style content domains

#* contextual spelling correction using latent semantic analysis
#@Michael P. Jones,James H. Martin
#t1997
#cApplied Natural Language Conferences
#index624
#id624
#%
#! contextual spelling errors defined use incorrect valid word particular sentence context traditional spelling checkers flag misspelled words typically attempt identify words used incorrectly sentence explore use latent semantic analysis correcting incorrectly used words results compared earlier work based bayesian classifier

#* maximum entropy approach adaptive statistical language modeling
#@Ronald Rosenfeld
#t1999
#cComputer, Speech and Language
#index625
#id625
#%
#! anadaptive statistical language model described successfully integrates long distance linguistic information knowledge sources existing statistical language models exploit immediate history text extract information documents history propose use trigger pairs basic information bearing elements allows model adapt expectations topic discourse statistical evidence multiple sources combined traditionally linear interpolation variants used shown seriously deficient instead apply principle maximum entropy information source gives rise set constraints imposed combined estimate intersection constraints set probability functions consistent information sources function highest entropy set solution given consistent statistical evidence unique solution guaranteed exist iterative algorithm exists guaranteed converge framework extremely general phenomenon described terms statistics text readily incorporated adaptive language model based approach trained wall street journal corpus showed perplexity reduction baseline interfaced sphinx ii carnegie mellons speech recognizer reduced error rate illustrates feasibility incorporating diverse knowledge sources single unified statistical framework

#* io index acces optimized query processing
#@Holger Bast,Debapriyo Majumdar,Ralf Schenkel,Martin Theobald,Gerhard Weikum
#t2006
#c Proceedings of VLDB'06
#index626
#id626
#%228,627,459,232,628,629,233,630,631,632,238,268,633,634,635,636,637,476,638,639,232,640,46,518,47,641,642,643
#! query processing important building block ranked retrieval applications ranging text data integration distributed aggregation network logs sensor data queries operate index lists querys elementary conditions aggregate scores result candidates best implementation methods setting family threshold algorithms aim terminate index scans early possible based lower upper bounds final scores result candidates procedure performs sequential disk accesses sorted index scans option performing random accesses resolve score uncertainty entails scheduling kinds accesses prioritization different index lists sequential accesses decision perform random accesses candidates prior literature studied scheduling issues access types isolation current paper takes integrated view scheduling issues develops novel strategies outperform prior proposals large margin main contributions new principled scheduling methods based knapsackrelated optimization sequential accesses cost model random accesses methods boosted harnessing probabilistic estimators scores selectivities index list correlations performance experiments different datasets trec terabyte http server logs imdb methods achieved significant performance gains compared best previously known methods

#* eddies continuously adaptive query processing
#@Ron Avnur,Joseph M. Hellerstein
#t2000
#cSIGMOD Conference
#index627
#id627
#%
#! large federated shared databases resources exhibit widely fluctuating characteristics assumptions time query submitted rarely hold duration query processing result traditional static query optimization execution techniques ineffective environments paper introduce query processing mechanism called eddy continuously reorders operators query plan runs characterize

#* effect adding relevance information relevance feedback environment
#@Chris Buckley,Gerard Salton,James Allan
#t1994
#cSIGIR
#index628
#id628
#%
#! effects adding information relevant documents examined trec routing environmentfia modified rocchio relevance feedback approach used varying number relevantdocuments retrieved initial smart search varying number terms relevantdocuments used expand initial queryfi recall precision evaluation reveals ofexpansion query adding terms relevant documents increases does effectivenessfi

#* efficient query calculation distributed networks
#@Pei Cao,Zhe Wang
#t2004
#cPODC
#index629
#id629
#%
#! width typically kb mb sfi enable web streaming media applications paper presents new algorithm answer queries efi gfi objects highest ag gregate values distributed networkfi existing algorithms threshold algorithm consume excessive bandwidth number nodes highfi propose new algorithm called phase uniform threshold tput fi tput reduces network bandwidth sumption pruning away ineligible objects ter minates round trips regardless data inputfi paper presents sets results tputfi trace driven simulations pending size network tput reduces network tra orders magnitude compared existing algorithmsfi second tput proven instance optimal data series sat isfy lower bound slope decreases valuesfi particular analysis shows using pruning parameter tput achieves qualitative duction network tra example lowering optimality ratio data series following zipf distributionfi

#* minimal probing supporting expensive predicates queries
#@Kevin Chen-Chuan Chang,Seung-won Hwang
#t2002
#cSIGMOD Conference
#index630
#id630
#%644,514,516,509,510,645,512,646,238,630,647,648,234,235,232
#! paper addresses problem evaluating ranked topfi queries expensive predicates major dbmss support expensive user defined predicates boolean queries believe support ranked queries important ranked queries need model user specific concepts preference relevance similarity dynamic user defined functions second middleware systems incorporate external predicates integrating autonomous sources typically accessible object queries fuzzy joins inherently expensive essentially user defined operations dynamically associate multiple relations predicates dynamically defined externally accessed rely index mechanisms provide zero time sorted output instead require object probe evaluate current standard sort merge framework ranked queries efficiently handle predicates completely probe objects sorting merging produce topfi answers minimize expensive probes develop formal principle necessary probes determines probe absolutely required propose algorithm mpro implementing principle provably optimal minimal probe cost mpro scale easily parallelized experiments using real estate benchmark database synthetic datasets mpro enables significant probe reduction orders magnitude faster standard scheme using complete probing

#* optimizing selection queries multimedia repositories
#@Surajit Chaudhuri,Luis Gravano,Amelie Marian
#t2004
#cIEEE Trans. Knowl. Data Eng.
#index631
#id631
#%
#! repositories multimedia objects having multiple types attributes image text increasingly common query attributes typically request just set objects traditional relational query model filtering grade match associated object indicates object matches selection condition ranking furthermore unlike relational model users just want ranked objects selection queries relatively small addition differences query model peculiarity multimedia repositories allow access attributes object indexes paper investigate optimize processing selection queries multimedia repositories access characteristics repositories query model lead novel issues query optimization particular choice indexes used search repository strongly influences cost processing filtering condition define execution space search minimal set indexes searched minimal general problem picking optimal plan search minimal execution space np hard present efficient algorithm solves problem optimally respect cost model execution space predicates query independent problem optimizing selection queries viewed cases evaluating traditional selection conditions problems viewed extended filtering problem techniques query processing optimization adapted

#* combining fuzzy information overview
#@Ronald Fagin
#t2002
#cSIGMOD Record
#index632
#id632
#%
#! assume object database grades scores attributesfi

#* efficient multi feature queries heterogeneous environments
#@Ulrich Guntzer,Wolf-Tilo Balke,Werner Kiesling
#t2001
#cITCC
#index633
#id633
#%
#! applications like multimedia databases enterprisewideinformation management systems meet thechallenge efficiently retrieving best matching objectsfrom vast collections data present new algorithmstream combine processing multi feature queries onheterogeneous data sourcesfi stream combine selfadaptingto different data distributions specifickind combining functionfi furthermore present anew retrieval strategy essentially speed

#* supporting join queries relational databases
#@Ihab F. Ilyas,Walid G. Aref,Ahmed K. Elmagarmid
#t2004
#cVLDB J.
#index634
#id634
#%231,232,233,234,630,649,512,238,463,633,650,651,652,647,518,653
#! ranking queries produce results ordered computed score typically queries involve joins users usually interested join results current relational query processors handle ranking queries efficiently especially joins involved paper address supporting join queries relational query processors introduce new rank join algorithm makes use individual orders inputs produce join results ordered userspecified scoring function idea rank join results progressively join operation introduce physical query operators based variants ripple join implement rankjoin algorithm operators non blocking integrated pipelined execution plans address practical issues optimization heuristics integrate new join operators practical query processors implement new operators inside prototype database engine based predator experimental evaluation approach compares recent algorithms joining ranked inputs shows superior performance

#* rank aware query optimization
#@Ihab F. Ilyas,Rahul Shah,Walid G. Aref,Jeffrey Scott Vitter,Ahmed K. Elmagarmid
#t2004
#cSIGMOD Conference
#index635
#id635
#%
#! ranking important property needs fully supported current relational query enginesfi recently rank join query operators proposed based rank aggregation algorithmsfi rank join operators progressively rank join results performing join operationfi new operators direct impact traditional query processing optimizationfiwe introduce rank aware query optimization framework fully integrates rank join operators relational query enginesfi framework based extending dynamic programming algorithm enumeration pruningfi define ranking interesting property triggers generation rank aware query plansfi unlike traditional join operators optimizing rank join operators depends estimating input cardinality operatorsfi introduce probabilistic model estimating input cardinality cost rank join operatorfi knowledge paper effort estimating needed input size optimal rank aggregation algorithmsfi costing ranking plans challenging key integration rank join operators real world query processing enginesfi experimentally evaluate framework modifying query optimizer open source database management systemfi experiments validity framework accuracy proposed estimation modelfi

#* history histograms
#@Yannis Ioannidis
#t2003
#cProc. of VLDB Conference
#index636
#id636
#%
#! history histograms long rich detailed information step includes course histograms different scientific fields successes failures histograms approximating compressing information adoption industry solutions given great variety histogram related problems paper spirit histogram techniques compress entire history including future history currently anticipated given fixed space budget recording details periods events results highest personally biased limited set experiments semantic distance compressed form history relatively small

#* integration structure indexes inverted lists
#@Raghav Kaushik,Rajasekar Krishnamurthy,Jeffrey F. Naughton,Raghu Ramakrishnan
#t2004
#cICDE
#index637
#id637
#%
#! methods proposed evaluate queries native xml dbms queries specify path keyword constraints broadly consist graph traversal approaches optimized auxiliary structures known structure indexes approaches based information retrieval style inverted lists propose strategy combines forms auxiliary indexes query evaluation algorithm branching path expressions based strategy technique general applicable wide range choices structure indexes inverted list join algorithms experiments niagara xml dbms benefit integrating forms indexes consider algorithmic issues evaluating path expression queries notion relevance ranking incorporated integrating techniques threshold algorithm proposed fagin et al obtain instance optimal algorithms push computation

#* ranksql query algebra optimization relational queries
#@Chengkai Li,Kevin Chen-Chuan Chang,Ihab F. Ilyas,Sumin Song
#t2005
#cSIGMOD Conference
#index638
#id638
#%
#! paper introduces ranksql provides atic principled framework support efficient evaluations ofranking queries relational database systems rdbms extending relational algebra query optimization previously query processing studied middleware scenario inrdbms piecemeal fashion focusing specific oper ator sitting outside core query engines contrast weaim support ranking class database construct akey insight new ranking relationship viewed anotherlogical property data parallel membership property ofrelational data model membership essentially supportedin rdbms support ranking clearly lacking ad dress fundamental integration ranking rdbms waysimilar membership boolean filtering supported extend relational algebra proposing rank relational modelto capture ranking property introducing new extendedoperators support ranking class construct enabled bythe extended algebra present pipelined incremental exe cution model ranking query plans expressed tra ditionally based fundamental ranking principle optimizetop queries propose dimensional enumeration algorithm toexplore extended plan space enumerating plans twodual dimensions ranking membership propose asampling based method estimate cardinality rank awareoperators costing plans experiments validity ofour framework accuracy proposed estimation model

#* optimized query execution large search engines global page ordering
#@Xiaohui Long,Torsten Suel
#t2003
#cVLDB
#index639
#id639
#%471,654,655,197,656,657,500,658,232,64,659,660,509,512,632,43,238,661,633,67,662,663,175,664,665,666,667,668,518,669,47,670,671,672,673,674,646,62,675
#! large web search engines answer thousands queries second interactive response times major factor cost executing query given lengths inverted lists query terms increase size document collection range megabytes address issue ir database researchers proposed pruning techniques compute approximate term based ranking functions scanning inverted lists years search engines incorporated new types ranking techniques exploit aspects hyperlink structure web popularity page obtain improved results focus question techniques efficiently integrated query processing particular study pruning techniques query execution large engines case global ranking pages provided pagerank method addition standard term based approach pruning schemes case evaluate efficiency experimental clusterbased search engine million web pages results significant potential benefit techniques

#* adaptive processing queries xml
#@Amelie Marian,Sihem Amer-Yahia,Nick Koudas,Divesh Srivastava
#t2005
#cICDE
#index640
#id640
#%
#! ability compute matches xml queries gaining importance increasing number large

#* efficient versatile query engine topx search
#@Martin Theobald,Ralf Schenkel,Gerhard Weikum
#t2005
#cVLDB
#index641
#id641
#%
#! paper presents novel engine coined topx efficient ranked retrieval xml documents semistructured nonschematic data collectionsfi algorithm follows paradigm threshold algorithms query processing focus inexpensive sequential accesses index lists judiciously scheduled random accesses difficulties applying

#* query evaluation probabilistic guarantees
#@Martin Theobald,Gerhard Weikum,Ralf Schenkel
#t2004
#cVLDB
#index642
#id642
#%
#! queries based ranking elements multidimensional datasets fundamental building block kinds information discoveryfi best known general purpose algo rithm evaluating queries fagin threshold algorithm ta fi user goal beh ind queries identify relevant novel data items intriguing use approximate variants ta reduce run time costsfi paper introduces family approximate algorithms based probabilistic argumentsfi scanning index lists lying multidimensional data space descending order local scores various forms convolution derived bounds employed predict safe high probability drop candidate items prune index scansfi precision efficiency developed methods experimentally evaluated based large web corpus structured data collectionfi fiintroduction fi motivation queries multidimensional datasets compute relevant interesting results partial match que ry based similarity scores attribute values gard elementary query conditions score aggrega tion function weighted summationfi funda mental building block information discovery arises important application classes web intranet search engines scores based word occur rence statistics possibly combining criteria like text based relevance link based authority recency multimedia similarity search feature vectors images music video preference queries structured semistructured data product catalogs custo mer support data having major text compo nent fi best known general purpose method queries fagin threshold algorithm known ta independently proposed ne pal et alfi ntzer et alfi fi method sumes attribute multidimensional data space index list access data items descending order local score

#* processing distributed queries
#@Hailing Yu,Hua-Gang Li,Ping Wu,Divyakant Agrawal,Amr El Abbadi
#t2005
#c16th Database and expert systems applications
#index643
#id643
#%
#! ranking aware queries queries received attention recently various contexts web multimedia retrieval relational databases distributed systems queries play critical role decision making related activities identifying interesting objects network monitoring load balancing paper study ranking aggregation problem distributed systems prior research addressing problem did data distributions account simply assuming uniform data distribution nodes realistic real data sets general inefficient paper propose efficient algorithms consider data distributions different ways extensive experiments demonstrate advantages approaches terms bandwidth consumption

#* framework expressing combining preferences
#@Rakesh Agrawal,Edward L. Wimmers
#t2000
#cSIGMOD Conference
#index644
#id644
#%
#! advent world wide web created explosion available line information range potential choices expand time effort required sort expands propose formal framework expressing combining user preferences address problem preferences used focus search queries order search results preference expressed user entity described set named fields field values certain type symbol used match element type set preferences combined using generic combine operator instantiated value function providing great deal flexibility preferences combined way combination preferences yields preference providing closure property demonstrate power framework illustrating currently popular personalization real life application realized special cases framework discuss implementation framework relational setting

#* efficient execution multi parametric ranked queries
#@Vagelis Hristidis,Nick Koudas,Y. Papakonstantinou,Yannis Papakonstantinou,La Jolla Ca,La Jolla Ca
#t2001
#cSIGMOD
#index645
#id645
#%
#! users need optimize selection objects appropriately weighting importance multiple object attributes optimization problems appear operations research applied mathematics everyday life buyer select home weighted function number attributes like distance office price area capture queries definition preference queries use weight function relations attributes derive score tuple database systems efficiently produce results preference query need evaluate weight function tuples relation prefer answers preference queries efficiently using materialized views preprocessed stored result preference query produced pipelined fashion using materialized view excellent performance delivered given reasonable number materialized views provide algorithm selects number views precompute materialize given space constraints implemented algorithms proposed paper prototype called prefer operates commercial database management present results performance comparison comparing algorithms prior approaches using synthetic datasets results indicate proposed algorithms superior performance compared approaches preprocessing preparation materialized views execution time

#* using fagins algorithm merging ranked results multimedia middleware
#@Edward Wimmers,Laura Haas,Mary Tork Roth,Christoph Braendli
#t1998
#cFourth IFCIS International Conference on Cooperative Information Systems
#index646
#id646
#%
#! distributed multimedia information allows applications access variety data different modalities stored data sources specialized search capabilities user request set objects ranked particular property combination properties fagin gives algorithm efficiently merging multiple ordered streams ranked results form new stream ordered combination ranks

#* supporting incremental join queries ranked inputs
#@Apostol Natsev,Yuan-Chi Chang,J.R. Smith,Chung-Sheng Li,Jeffrey Scott Vitter
#t2001
#cVLDB
#index647
#id647
#%
#! paper investigates problem incrementaljoins multiple ranked data sets joincondition list arbitrary user defined predicateson input tuplesfi problem arises inmany important applications dealing orderedinputs multiple ranked data sets requiringthe solutionsfi use multimedia applicationsas motivating examples problemis equally applicable traditional database applicationsinvolving optimal resource

#* saying sql
#@Michael J. Carey,Donald Kossmann
#t1997
#cACM SIGMOD Record
#index648
#id648
#%
#! paper study simple sql extension enables query writers explicitly limit cardinality query result examine impact query optimization run time execution components relational dbms presenting approachesa conservative approach aggressive approachto exploiting cardinality limits relational query plans results obtained empirical study conducted using db demonstrate benefits sql extension illustrate tradeoffs approaches implementing

#* rank aggregation methods web
#@Cynthia Dwork,Ravi Kumar,Moni Naor,D. Sivakumar
#t2001
#cWWW
#index649
#id649
#%
#! consider problem combining ranking results fromvarious sourcesfi context web main applicationsinclude building meta search engines combiningranking functions selecting documents based multiplecriteria improving search precision word associationsfiwedevelop set techniques rank aggregationproblem compare performance ofwell known methodsfi primary goal work designrank aggregation techniques

#* ripple joins online aggregation
#@Peter J. Haas,Joseph M. Hellerstein
#t1999
#cSIGMOD Conference
#index650
#id650
#%
#! present new family join algorithms called ripple joins online processing multi table aggregation queries rela tional database management dbms queries arise naturally interactive exploratory decision support applications traditional offline join algorithms designed minimize time completion query contrast ripple joins designed minimize time acceptably precise esti mate query result available measured length confidence interval ripple joins adaptive adjusting behavior processing accordance statis tical properties data ripple joins permit user dynamically trade key performance factors line aggregation time successive updates run ning aggregate confidence interval length decreases update ripple joins implemented existing dbms using iterators overview methods used compute confidence intervals adaptively optimize ripple join aspect ratio parameters experiments initial implementation algorithms postgres dbms time required produce reasonably precise online estimates orders magnitude smaller time required best offline join algorithms pro duce exact answers

#* optimization parallel query execution plans xprs
#@Wei Hong,Michael Stonebraker
#t1993
#cDistributed and Parallel Databases
#index651
#id651
#%
#! authors approach optimization query execution plans xprs multi user parallel database machine based shared memory multi processor disk array main difficulties optimization problem compile time unknown parameters available buffer size number free processors enormous search space possible parallel plans authors deal problems novel phase optimization strategy dramatically reduces search space allows run time parameters significantly compromising plan optimality present phase strategy experimental evidence xprs benchmarks indicate produces optimal plans

#* joining ranked inputs practice
#@Ihab F. Ilyas,Walid G. Aref,Ahmed K. Elmagarmid
#t2002
#cVLDB
#index652
#id652
#%
#! joining ranked inputs essential requirement database applications ranking search results multiple search engines answering multi feature queries multimedia retrieval systems introduce new practical pipelined query operator termed nra rj produces global rank input ranked streams based score function output nra rj serve valid input nra rj operators query pipeline nra rj operator support

#* xjoin reactively scheduled pipelined join operator
#@Tolga Urhan University,Tolga Urhan,Michael J. Franklin
#t2000
#cIEEE Data Engineering Bulletin
#index653
#id653
#%
#! wide area distribution raises significant performance problems traditional query processing techniques data access predictable link congestion load imbalances temporary outages pipelined query execution promising approach coping unpredictability environments allows scheduling adjust arrival properties data developed non blocking join operator called xjoin small memory footprint

#* searching web
#@Arvind Arasu,Junghoo Cho,Hector Garcia-Molina,Andreas Paepcke,Sriram Raghavan
#t2001
#cACM Trans. Internet Techn.
#index654
#id654
#%
#! offer overview current web search engine design introducing generic search enginearchitecture examine engine component turn cover crawling local web page storage indexing use link analysis boosting search performance common design andimplementation techniques components presented draw presentationfrom literature experimental search engine testbed emphasis introducingthe fundamental concepts results performance analyses conducted comparedifferent designs

#* distributed query processing using partitioned inverted files
#@Claudine Santos Badue,Ricardo A. Baeza-Yates,Berthier A. Ribeiro-Neto,Nivio Ziviani
#t2001
#cSPIRE
#index655
#id655
#%
#! paper study query processing distributed text database novelty real distributed architec ture implementation offers concurrent query servicefi distributed adopts network workstations model client server paradigmfi document col lection indexed inverted filefi adopt dis tinct strategies index partitioning distributed sys tem local index partitioning global index parti tioningfi strategies documents ranked using vector space model document filtering tech nique fast rankingfi evaluate compare impact index partitioning strategies query processing performancefi experimental results retrieval efficiency framework global index parti tioning outperforms local index partitioningfi

#* finding authorities hubs link structures world wide web
#@Allan Borodin,Gareth O. Roberts,Jeffrey S. Rosenthal,Panayiotis Tsaparas
#t2001
#cWWW
#index656
#id656
#%
#! recently anumberofalgorh prc osedforanalyzing hyperdcc linkstr tdsoastodetercdthebest authorodhum fora given topicorquer fi suchanalysis usually combined content analysis ther isa sense somealgoruo ar deemed morbalanced andother morfocused fiweunderdea compar study hyperdtd link analysisalgor cdco guidedby experd talquer cdwe prd ose somefor crc orforevaluating andcompartlinkanalysisalgor cuh

#* lessons giant scale services
#@Eric A. Brewer
#t2001
#cIEEE Internet Computing
#index657
#id657
#%
#! web portals isps aol microsoft network yahoo grown tenfold past years despite scale growth rates rapid evolution content features sites giant scale services like instant messaging napster available major web sites ebay cnn wal mart similar availability requirements article look basic model services focusing key realworld challenges face high availability evolution growth developing principles attacking problems

#* resemblance containment documents
#@AZ Broder
#t1997
#cSEQUENCES
#index658
#id658
#%
#! given documents ne mathematical notions resemblance containment capture informal notions roughly roughly containedfi basic idea reduce issues set intersection problems easily evaluated process random sampling inde pendently documentfi furthermore resemblance evaluated using xed size sample documentfi paper discusses mathematical properties measures cient implementation sampling process using rabin ngerprintsfi

#* evaluating performance distributed architectures information retrieval using variety workloads
#@Brendon Cahoon,Kathryn S. McKinley,Zhihong Lu
#t2000
#cACM Trans. Inf. Syst.
#index659
#id659
#%
#! information explosion internet elswhere offers access increasing number document collections order users effectively access collections information retrieval ir systems provide coordinated concurrent distributed access article explore achieve scalable performance distributed collection sizes ranging gb gb implement fully functional distributed ir based multithreaded version inquery simulation model measure performance function parameters client command rate number document collections ter ms query query term frequency number answers returned command mixture results important model query document commands heterogeneity commands significantly impacts performance based results recommend simple changes prototype evaluate changes using simulator significant resource demands information retrieval difficult generate workloads overwhelm resources regardless architecture realistic workloads demonstrate organizations response time gracefully degrades workload increases performance scales number processors scalable architecture includes surprisingly small number brokers large number clients servers communicate

#* automatic resource list compilation analyzing hyperlink structure associated text
#@Soumen Chakrabarti,Byron Dom,Prabhakar Raghavan,S. Rajagopalan,D. Gibson,J. Keinberg
#t1998
#cPROCEEDINGS OF THE SEVENTH INTERNATIONAL WORLD WIDE WEB CONFERENCE
#index660
#id660
#%
#! design prototyping evaluation arc automatically compiling list authoritative web resources sufficiently broad topic goal arc compile resource lists similar provided yahoo infoseek fundamental difference services construct lists manually combination human automated effort arc operates fully automatically evaluation arc yahoo infoseek resource lists panel human users evaluation suggests resources arc frequently fare better lists resources manually compiled classified topic provide examples arc resource lists reader examine

#* optimizing multifeature queries image databases
#@Ulrich Guntzer,Wolf-Tilo Balke,Werner Kiesling
#t2000
#cProc. of the Intern. Conf. on Very Large Databases VLDB 2000
#index661
#id661
#%
#! digital libraries image retrieval queries based similarity objects using feature attributes like shape texture color text multi feature queries return ranked result set instead exact matches user wants ranked objects present new algorithm called quick combine european patent pending nr ep combining multi feature result lists guaranteeing correct retrieval ranked results score aggregation virtually combining function used including weighted queries compared fagins algorithm developed improved termination condition tuned combination heuristic control flow adopting itself narrowly particular score distribution ranked results computed output incrementally dramatically improve performance particular non uniform score distributions benchmarks practical data indicate efficiency gains factor skewed data observed speed factors larger performance results scale different database sizes numbers result sets combine

#* topic sensitive pagerank
#@T. Haveliwala
#t2002
#cWWW
#index662
#id662
#%
#! original pagerank algorithm improving rankingof search query results single pagerank vector computed using link structure web capture therelative importance web pages independent particularsearch queryfi yield accurate search results propose computing set pagerank vectors biased usinga set representative topics capture accuratelythe notion importance respect particular topicfiby using

#* inverted file partitioning schemes multiple disk systems
#@Byeong-Soo Jeong,Edward Omiecinski
#t1995
#cIEEE Trans. Parallel Distrib. Syst.
#index663
#id663
#%
#! multiple disk systems disk arrays attractive approach meet highperformance demands data intensive applications information retrieval systemsfiwhen partition distribute files multiple disks exploit potential oparallelism balanced workload distribution important good performancefinaturally performance parallel information retrieval using inverted filestructure affected partitioning

#* stochastic approach link structure analysis salsa tkc effect
#@Ronny Lempel,Shlomo Moran
#t2000
#cComputer Networks
#index664
#id664
#%
#! today searching information www usually performsa query term based search engine engine return thequery result list web sites contents matches queryfi forbroad topic queries searches result huge set retrieveddocuments irrelevant userfi informationis contained link structure wwwfi information aswhich pages linked used augment

#* optimizing result prefetching web search engines segmented indices
#@Ronny Lempel,Shlomo Moran
#t2004
#cACM Trans. Internet Techn.
#index665
#id665
#%
#! study process search engineswith segmented indices serve queries par ticular investigate number resultpages search engines prepareduring query processing phase search engine users observed tobrowse pages results forqueries submit behavior ofusers suggests prefetching resultsupon processing initial query cient prefetched results willnot requested user initiatedthesearch policy abandons sult prefetching favor retrieving just therst page search results mightnot makeop timal use resources argue certain behavior users engines prefetch constant numberof result pages query dene crete query processing model search en gines segmented indices analyze thecost prefetching policies based onthese costs determine constant optimizes prefetching pol icy results applicable tolocalindexpartitions inverted les arealso applicable processing short queriesinglobal indexarchitectures

#* scalable distributed architectures information retrieval
#@Zhihong Lu,Kathryn S. McKinley
#t1999
#cElectronic Doctoral Dissertations for UMass Amherst
#index666
#id666
#%
#! information explodes internet intranets information retrieval ir systems cope challenge scale provide scalable performance rapidly increasing data workloads critical design generation information retrieval systems dissertation studies scalable distributed ir architectures provide quick response maintain acceptable retrieval accuracy distributed architectures exploit parallelism information retrieval cluster parallel ir servers using symmetric multiprocessors use partial collection replication selection collection selection restrict search small percentage data maintaining retrieval accuracy investigate using partial collection replication ir systems examine query locality real systems select partial replica based relevance load balance replicas original collection updating overheads strategies results exists sufficient query locality justify partial replication information retrieval proposed replica selection algorithm effectively selects relevant partial replicas inexpensive implement evidence indicates partial replication achieves better performance caching queries replica selection algorithm finds similarity non identical queries increases observed locality use validated simulator perform detailed performance evaluation distributed ir architectures explore best build parallel ir servers using symmetric multiprocessors evaluate performance partial collection replication collection selection compare performance partial collection replication collection partitioning collection selection

#* building distributed text index web
#@Sergey Melnik,Sriram Raghavan,Beverly Yang,Hector Garcia-Molina
#t2001
#cWWW
#index667
#id667
#%
#! identify crucial design issues building distributed inverted index large collection web pages introduce novel pipelining technique structuring core index building substantially reduces index construction time propose storage scheme creating managing inverted files using embedded database suggest compare different strategies collecting global statistics distributed inverted indexes finally present performance results experiments testbed distributed web indexing implemented

#* breadth search crawling yields high quality pages
#@Marc Najork,Janet L. Wiener
#t2001
#cProc. 10th International World Wide Web Conference
#index668
#id668
#%
#! paper examines average page quality time pages downloaded web crawl million unique pagesfi use connectivity based metric pagerank measure quality pagefi traversing web graph breadth search order good crawling strategy tends discover high quality pages early crawlfi

#* pagerank citation ranking bringing order web
#@Lawrence Page,Sergey Brin,Rajeev Motwani,Terry Winograd
#t1999
#cStanford InfoLab
#index669
#id669
#%
#! importance webpage inherently subjective matter depends readers interests knowledge attitudes said objectively relative importance web pages paper describes pagerank method rating web pages objectively mechanically effectively measuring human attention devoted compare pagerank idealized random websurfer efficiently compute pagerank large numbers pages apply pagerank search user navigation

#* intelligent surfer probabilistic combination link content information pagerank
#@Matthew Richardson,Pedro Domingos
#t2001
#cNIPS
#index670
#id670
#%
#! pagerank algorithm used google search engine greatly improves results web search taking account link structure webfi pagerank assigns page score propor tional number times random surfer visit page surfed indefinitely page page following outlinks page equal probabilityfi propose improve page rank using intelligent surfer guided probabilistic model relevance page queryfi efficient execution algorithm query time possible pre computing crawl time queries neces sary termsfi experiments large subsets web indicate algorithm significantly outperforms pagerank hu man rated quality pages returned remaining efficient used today large search enginesfi

#* search engines web dynamics
#@Knut Magne Risvik,Rolf Michelsen
#t2002
#cComputer Networks
#index671
#id671
#%
#! paper study dimensions web dynamics context large scale internetsearch enginesfi growth update dynamics clearly represent big challenges searchenginesfi problems arise components reference search engine modelfi

#* design implementation high performance distributed web crawler
#@Vladislav Shkapenyuk,Torsten Suel
#t2002
#cICDE
#index672
#id672
#%
#! broad web search engines specialized search tools rely webcrawlers acquire large collections pages indexing analysisfi web crawlermay interact millions hosts period weeks months issues robustness flexibility manageability major importancefi addition performance network resources os limits taken account order achieve high performanceat reasonable costfiin

#* odissea peer peer architecture scalable web search information retrieval
#@Torsten Suel,Chandan Mathur,Jo-wen Wu,Jiangong Zhang,Alex Delis,Mehdi Kharrazi,Xiaohui Long,Kulesh Shanmugasundaram
#t2003
#cWebDB
#index673
#id673
#%
#! consider problem building based search engine massive document collections prototype called odissea open distributed search engine architecture currently development group odissea provides highly distributed global indexing query execution service used content residing inside outside network odissea different approaches search assumes tier search engine architecture global index structure distributed network overview proposed discuss basic design choices main focus efficient query execution discuss recent work queries database community applied highly distributed environment preliminary simulation results real search engine log terabyte web collection indicate good scalability approach

#* performance inverted indices distributed text document retrieval systems
#@Anthony Tomasic,Hector Garcia-Molina
#t1993
#cPDIS
#index674
#id674
#%
#! performance distributed text document retrieval systems strongly influenced bythe organization inverted indexfi paper compares performance impact queryprocessing various physical organizations inverted listsfi present new probabilisticmodel database queriesfi simulation experiments determine variables moststrongly influence response time throughputfi leads set design trade offs awide range hardware

#* implementations partial document ranking using inverted files
#@Wai Yee Peter Wong,Dik Lun Lee
#t1993
#cInf. Process. Manage.
#index675
#id675
#%
#! commercial text retrieval systems employ inverted files improve retrieval speedfithis paper concerns implementations document ranking based inverted filesfi threeheuristic methods implementing tf thetaidf weighting strategy tf stands term frequencyand idf stands inverse document frequency studiedfi basic idea heuristic methodsis process query terms order documents possible identified

#* missing linka probabilistic model document content hypertext connectivity
#@David Cohn,Thomas Hofmann
#t2001
#cNIPS 13
#index676
#id676
#%
#! joint probabilistic model modeling contents inter connectivity document collections sets web pages research paper archives model based probabilistic factor decomposition allows identifying principal topics collection authoritative documents topics furthermore relationships topics mapped order build predictive model link content applications approach information retrieval search topic identification query disambiguation focused web crawling web authoring bibliometric analysis

#* image information retrieval overview current research
#@A. Goodrum
#t2000
#cInforming Science
#index677
#id677
#%
#! paper provides overview current research image information retrieval provides outline areas future researchfi approachis broad interdisciplinary focuses aspects image research ir text based retrieval content based retrieval userinteractions image information retrieval systemsfi review concludes image retrieval evaluation studies similar trecfikeywords information science image retrieval cbir

#* automatic image annotation retrieval using cross media relevance models
#@Jiwoon Jeon,Victor Lavrenko,R. Manmatha
#t2003
#cSIGIR
#index678
#id678
#%
#! libraries traditionally used manual image annotationfor indexing later retrieving image collectionsfihowever manual image annotation expensive laborintensive procedure great interestin coming automatic ways retrieve imagesbased contentfi propose automatic approachto annotating retrieving images based training setof imagesfi assume regions image describedusing small vocabulary

#* introduction variational methods graphical models
#@Michael I. Jordan,Zoubin Ghahramani,Tommi Jaakkola,Lawrence K. Saul
#t1999
#cMachine Learning
#index679
#id679
#%
#! paper presents tutorial introduction use variational methods inference learning graphical models bayesian networks markov random fields present number examples graphical models including qmr dt database sigmoid belief network boltzmann machine variants hidden markov models infeasible run exact inference algorithms introduce variational methods exploit laws large numbers transform original graphical model simplified graphical model inference efficient inference simpified model provides bounds probabilities original model general framework generating variational transformations based convex duality finally return examples demonstrate variational algorithms formulated case

#* model multimedia information retrieval
#@Carlo Meghini,Fabrizio Sebastiani,Umberto Straccia
#t2001
#cJ. ACM
#index680
#id680
#%
#! research multimedia information retrieval mir recently witnessed booming interestfi prominent feature research trend simultaneous independent materialization fields sciencefi resulting richness paradigms methods systems long run result fragmentation efforts slow progressfi primary goal study promote integration methods techniques mir contributing conceptual model encompasses unified coherent perspective efforts produced label mirfi model offers retrieval capability spans media text images dimensions form content structurefi way reconciles similarity based methods semantics based ones providing guidelines design systems able provide generalized multimedia retrieval service existing forms retrieval coexist combined desired mannerfi model formulated terms fuzzy description logic plays twofold role directly models semantics based retrieval offers ideal framework integration multimedia multidimensional aspects retrieval mentioned abovefi model accounts relevance feedback text image retrieval integrating known techniques taking account user judgmentsfi implementation model addressed presenting decomposition technique reduces query evaluation processing simpler requests solved means widely known methods text image retrieval semantic processingfi prototype multidimensional image retrieval presented shows decomposition technique work significant casefi

#* probabilistic framework semantic video indexing filtering retrieval
#@H.Naphide,T.Huang
#t2001
#cMultimedia, IEEE Transactions on
#index681
#id681
#%
#! semantic filtering retrieval multimedia content crucial efficient use multimedia data repositories video query semantic keywords difficult problems multimedia data retrieval difficulty lies mapping low level video representation high level semantics formulate multimedia content access problem multimedia pattern recognition problem propose probabilistic framework semantic video indexing support filtering retrieval facilitate efficient content based access map low level features high level semantics propose probabilistic multimedia objects multijects examples multijects movies include explosion mountain beach outdoor music semantic concepts videos interact model interaction explicitly propose network multijects multinet using probabilistic models site multijects rocks sky snow water body forestry greenery outdoor using bayesian belief network multinet demonstrate application framework semantic indexing demonstrate detection performance significantly improved using multinet interconceptual relationships account multinet fuse heterogeneous features support detection based inference reasoning

#* estimating dirichlet distribution
#@Thomas P. Minka
#t2003
#cnull
#index682
#id682
#%
#! dirichlet distribution compound variant dirichlet multinomial basic models proportional data mix vocabulary words text document maximum likelihood estimate distributions available closed form paper describes simple efficient iterative schemes obtaining parameter estimates models case fixed point iteration newton raphson generalized newton raphson iteration provided

#* modeling predicting personal information dissemination behavior
#@Xiaodan Song,Ching-Yung Lin,Belle L. Tseng,Ming-Ting Sun
#t2005
#cKDD
#index683
#id683
#%684,685,686,687,688,689,690,669,175,691,692,568,187,182,371,444,693,694,230
#! paper propose new way automatically model predict human behavior receiving disseminating information analyzing contact content personal communications personal profile called communitynet established individual based novel algorithm incorporating contact content time information simultaneously used personal social capital management clusters communitynets provide view informal networks organization management new algorithm developed based combination dynamic algorithms social network field semantic content classification methods natural language processing machine learning literatures tested communitynets enron email corpus report experimental results including filtering prediction recommendation capabilities personal behavior intention somewhat predictable based models instance person going send specific email predicted ones personal social network content analysis experimental results prediction accuracy proposed adaptive algorithm better social network based predictions better aggregated model based latent dirichlet allocation social network enhancement online demo systems developed allow interactive exploration communitynet discussed

#* know know work information age
#@BA Nardi,S Whittaker,H Schwarz
#t2000
#cFirst Monday
#index684
#id684
#%
#! discuss ethnographic research personal social networks workplace arguing traditional institutional resources replaced resources workers networks social networks key sources labor information rapidly transforming economy characterized institutional stability fewer reliable corporate resources personal social network fast sensible alternative traditional org chart everyday transactions today economy

#* structure culture simmelian ties entrepreneurial firms
#@David Krackhardt,Martin Kilduff
#t2002
#cSocial Networks
#index685
#id685
#%
#! article develops cultural agreement approach organizational culture emphasizes clusters individuals reinforce potentially idiosyncratic understandings aspects culture including structure network relations building recent work concerning simmelian tied dyads defined dyads embedded person cliques research examines perceptions concerning advice friendship relations entrepreneurial firms results support idea simmelian tied dyads relative dyads general reach higher agreement concerning tied embedded triads organizations

#* contactmap integrating communication information visualizing personal social networks
#@BA Nardi,S Whittaker,Ellen Isaacs,Mike Creech,Jeff Johnson,John Hainsworth
#t2001
#cCommunications of the ACM
#index686
#id686
#%
#! short years witnessed massive uptake use cell phones personal digital assistants hybrid devices integrate phone internet services devices communicate traditional computers internet transformed computational environments communication spaces goal research seamlessly integrate communication traditional information functions computational devices provide organizing principle advanced user interfaces afford information communication services single integrated based research communication patterns workplace answer models personal social networks research showed people invest considerable effort maintaining links networks colleagues acquaintances friends networks significant organizing principle work information article report study workplace communication informed development efforts evolving software prototype contactmap user experiments contactmap

#* discovering shared interests people using graph analysis
#@Michael F. Schwartz,David C. M.Wood
#t1993
#cCommunications of the ACM
#index687
#id687
#%
#! important problem faced users large networks discover resources data orpeople paper focus locating people particular interests expertise usual approach tobuild group lists explicitly registered data doing assumes knows lists bebuilt included list present alternative approach support finegrained dynamically adaptive notion shared interests approach deduces interests history ofelectronic mail communication using set heuristic graph algorithms demonstrate algorithms apply ing data collected sites months using algorithms able deduce sharedinterest lists people far data collection sites algorithms present powerful abusedcan threaten privacy propose guidelines believe underly ethical use algorithms wediscuss possible applications believe threaten privacy including discovering resources otherthan people file data

#* referral web combining social networks collaborative filtering
#@Henry A. Kautz,Bart Selman,Mehul A. Shah
#t1997
#cCommun. ACM
#index688
#id688
#%
#! paper appears communications acm volfi nofi march finumerous studies shown effective channelsfor dissemination information expertise organization itsinformal network collaborators colleagues friends granovetter kraut wasserman galaskiewicz fi social network important official organizational structure tasks rangingfrom immediate local

#* self organization identification web communities
#@Gary William Flake,S Lawrence,C. Lee Giles,Frans Coetzee
#t2002
#cIEEE Computer
#index689
#id689
#%
#! despite decentralized unorganized nature web web self organizes communities highly related pages efficiently identified based purely connectivityfi

#* email spectroscopy automated discovery community structure organizations
#@JR Tyler,Dennis M. Wilkinson,BA Huberman
#t2003
#cThe Information Society
#index690
#id690
#%
#! method automatic identification communities practice email logs organizationfi use betweenness centrality algorithm rapidly communities graph representing information flowsfi apply algorithm email corpus nearly million messages collected month span method effective identifying true communities formal informal scale free graphsfi approach enables identification leadership roles communitiesfi studies complemented qualitative evaluation results fieldfi

#* models longitudinal network data
#@Tom Snijders Ics,Tom A.B. Snijders
#t2005
#cModels and Methods in Social Network Analysis
#index691
#id691
#%
#! chapter treats statistical methods network evolutionfi itis argued fruitful consider models networkevolution represented result usually non observed small changes occurring consecutively observed networksfi

#* link prediction problem social networks
#@David Liben-Nowell,Jon M. Kleinberg
#t2003
#cCIKM
#index692
#id692
#%
#! given snapshot social network infer new interactions membersare likely occur near future formalize question link prediction problem develop approaches link prediction based measures analyzing proximity ofnodes networkfi experiments large authorship networks suggest informationabout future interactions extracted network topology fairly subtlemeasures detecting

#* expertisenet relational evolutionary expert modeling
#@Xiaodan Song,Belle L. Tseng,Ching-Yung Lin,Ming-Ting Sun
#t2005
#cUser Modeling
#index693
#id693
#%
#! develop novel user centric modeling technology dynamically update person expertise profile enterprise environment technology enhance employees collaboration productivity assisting finding experts training employees instead using traditional search methods keyword match propose use relational evolutionary graph models expertisenet experts expertisenets used mining retrieval visualization conduct experiments building expertisenets researchers research paper collection experiments demonstrate expertise mining matching efficiently achieved based proposed relational evolutionary graph models

#* line new event detection tracking
#@James Allan,Ron Papka,Victor Lavrenko
#t1998
#cSIGIR
#index694
#id694
#%
#! work discuss evaluate solutions text classification problems associated events reported line source news present solutions related classification problems new event detection event clustering event tracking primary focus thesis new event detection goal identify news stories previously reported stream broadcast news comprising radio television newswire present algorithm new event detection analyze effects incorporating domain properties classification algorithm explore solution models temporal relationship news stories investigate use proper noun phrase extraction capture contained news results new event detection suggest previous approaches document clustering provide good basis approach new event detection improvements classification accuracy obtained domain properties broadcast news modeled new event detection related problem event clustering goal group stories discuss event investigate line clustering approach new event detection evaluate existing cluster comparison strategies previously used document retrieval results suggest strategies produce different groupings events line single link strategy extended model domain properties faster effective approaches dissertation explore test representation issues context event tracking classifier event formulated sample stories classifier used monitor subsequent news stream documents related event

#* multiscale topic tomography
#@Ramesh Nallapati,Susan Ditmore,J Lafferty,Kin Ung
#t2007
#cKDD
#index695
#id695
#%346,182,442,551,349,696,209,697,698,699,700,433,701
#! modeling evolution topics time great value automatic summarization analysis large document collections work propose new probabilistic graphical model address issue new model multiscale topic tomography model mttm employs non homogeneous poisson processes model generation word counts evolution topics modeled multi scale analysis using haar wavelets new features model modeling evolution topics various time scales resolution allowing user zoom time scales experiments science data using new model uncovers interesting patterns topics new model comparable lda predicting unseen data demonstrated perplexity experiments

#* probabilistic approach automatic keyword indexing
#@Stephen P Harter
#t196
#cJournal of the American Society for Information Science
#index696
#id696
#%
#! confirms previously published research concluding specialty words tend possess frequency distributions described single poisson distribution

#* bayesian multiscale models poisson processes
#@Eric D. Kolaczyk
#t1994
#cJournal of the American Statistical Association
#index697
#id697
#%
#! focus article new class models bayesian multi scale models bmsm phenomena defining characteristics observed data modeled discrete signal time series independent poisson counts underlying intensity function potentially structure multiple scales fundamental analysis data estimation underlying intensity function problem particular addressed knowledge intensity goal itself serve initial step prior higher level analyses detection classification

#* multiscale hidden markov models bayesian image analysis
#@Robert D. Nowak
#t1999
#cICIP99
#index698
#id698
#%
#! bayesian multiscale image analysis weds powerful modelingframework probabilistic graphs intuitively appealing andcomputationally tractable multiresolution paradigmfi addition providinga natural useful framework modeling processingimages bayesian multiscale analysis computationally demandingcompared classical markov random field modelsfi chapterfocuses probabilistic graph model called multiscale hidden markovmodel

#* statistical multiscale framework poisson inverse problems
#@Robert D. Nowak,Eric D. Kolaczyk
#t2000
#cIEEE Transactions on Information Theory
#index699
#id699
#%
#! paper describes statistical modeling analysis method linear inverse problemsinvolving poisson data based novel multiscale frameworkfi framework itself foundedupon multiscale analysis associated recursive partitioning underlying intensity acorresponding multiscale factorization likelihood induced analysis choiceof prior probability distribution match factorization modeling splits inthe underlying

#* simple effective approximations poisson model probabilistic weighted retrieval
#@S.E. Robertson,S. Walker
#t1994
#cSIGIR
#index700
#id700
#%
#! poisson model term frequencies used suggest ways incorporatingcertain variablesin probabilisticmodels informationretrieval variables concerned documenttermtkequency document length queryterm frequency simple weighting functionsare devel oped tested trec test collection considerable performance improvements simpleinverse collection frequency weighting demonstrated

#* topics time non markov continuous time model topical trends
#@Xuerui Wang,A McCallum
#t2006
#cKDD
#index701
#id701
#%182,702,371,349,703,704,705,569,338,444,436,683,706,558
#! paper presents lda style topic model captures low dimensional structure data structure changes time unlike recent work relies markov assumptions discretization time topic associated continuous distribution timestamps generated document mixture distribution topics influenced word occurrences documents timestamp meaning particular topic relied constant topics occurrence correlations change significantly time present results months personal email years nips research papers years presidential state union addresses showing improved topics better timestamp prediction interpretable trends

#* mixed membership models scientific publications
#@Elena Erosheva Stephen,Stephen Fienberg,J Lafferty
#t2004
#cProceedings of the National Academy of Sciences
#index702
#id702
#%
#! pnas world cited multidisciplinary scientific journals pnas official classification structure subjects reflected topic labels submitted authors articles largely related traditionally established disciplines include broad field classifications physical sciences biological sciences social sciences subtopic classifications fields focusing biological sciences explore internal soft classification structure articles based semantic decompositions bibliographies compare formal discipline classifications model assumes fixed number internal categories characterized multinomial distributions words abstracts references bibliographies soft classification article based proportions article content coming category discuss appropriateness model pnas database features data relevant soft classification

#* bursty hierarchical structure streams
#@Jon M. Kleinberg
#t2003
#cData Min. Knowl. Discov.
#index703
#id703
#%
#! fundamental problem text data mining extract meaningful structure document streams arrive continuously time mail news articles natural examples streams characterized topics appear grow intensity period time fade away published literature particular research field seen exhibit similar phenomena longer time scale underlying text mining work area following intuitive premise appearance topic document stream signaled burst activity certain features rising sharply frequency topic emerges goal present work develop formal approach modeling bursts way robustly efficiently identified provide organizational framework analyzing underlying content approach based modeling stream using infinite state automaton bursts appear naturally state transitions ways viewed drawing analogy models queueing theory bursty network traffic resulting algorithms highly efficient yield nested representation set bursts imposes hierarchical structure overall stream experiments mail research paper archives suggest resulting structures natural meaning terms content gave rise

#* generalized probability density function double bounded random processes
#@P.Kumaraswamy
#t1980
#cJournal of Hydrology
#index704
#id704
#%
#! author developed sinepower probability density function sp pdf fit random processes bounded lower upper ends mode occurring bounds condition relaxed generalized pdf entitled double bounded probability density function db pdf developed methods application practical problems parameter estimation simulation random variables explained numerical example author developed sinepower probability density function sp pdf fit random processes bounded lower upper ends mode occurring bounds condition relaxed generalized pdf entitled double bounded probability density function db pdf developed methods application practical problems parameter estimation simulation random variables explained numerical example

#* modeling word burstiness using dirichlet distribution
#@Rasmus Elsborg Madsen,David Kauchak,Charles Elkan
#t2005
#cICML
#index705
#id705
#%
#! multinomial distributions used model text documents capture phenomenon words document tend appear bursts word appears likely appear paper propose dirichlet compound multinomial model dcm alternative multinomial dcm model additional degree freedom allows capture burstiness experimentally dcm substantially better multinomial modeling text data measured perplexity using standard document collections dcm leads better classification multinomial model dcm performance comparable obtained multiple heuristic changes multinomial model

#* timemines constructing timelines statistical models word usage
#@Russell Swan,David Jensen
#t2000
#cACM SIGKDD 2000 Workshop on Text Mining
#index706
#id706
#%
#! present timemines automatically generatestimelines date tagged free text corporafi timeminesdetects ranks groups semantic features based ontheir statistical propertiesfi use features discoversets related stories deal single topicfi

#* opinion extraction semantic classification product reviews
#@K Dave,S Lawrence,DM Pennock
#t2003
#cInternational World Wide Web Conference
#index707
#id707
#%708,709,406,710,408,409,614,711,326,712,713,714,149,428,715,716,717,718,719,420,720,721,722
#! web contains wealth product reviews sifting daunting task ideally opinion mining tool process set search results given item generating list product attributes quality features aggregating opinions poor mixed good begin identifying unique properties problem develop method automatically distinguishing positive negative reviews classifier draws information retrieval techniques feature extraction scoring results various metrics heuristics vary depending testing situation best methods work better traditional machine learning operating individual sentences collected web searches performance limited noise ambiguity context complete web based tool aided simple method grouping sentences attributes results qualitatively quite useful

#* yahoo amazon sentiment parsing small talk web
#@SR Das,MY Chen
#t2001
#cEFA 2001 Barcelona Meetings
#index708
#id708
#%
#! internet feasible tap continuous stream public sentiment world wide web quite literally permitting feel pulse issue consideration present methodology real time sentiment extraction domain finance advent web sharp increase influence individuals stock market web based trading posting sentiment stock message boards important capture sentiment small investors index sentiment compiled paper comprises technology extracting small investor sentiment web sources create index illustrative applications methodology make use computerized natural language statistical algorithms automated classification messages posted web design suite classification algorithms different theoretical content view characterizing sentiment single posting message board use multiple methods allows imposition voting rules classification process enables elimination fuzzy messages better uninterpreted majority rule algorithms vastly improves classification accuracy leads natural increase number messages classified fuzzy classifier achieves accuracy versus random classification accuracy compares favorably human agreement message classification technology computationally efficient allowing access interpretations thousands messages minutes illustrative applications evidence strong link market movements sentiment based approximately messages quarter evidence sentiment based stock movements

#* word grams
#@Fernando C. N. Pereira,Yoram Singer,Naftali Tishby
#t1996
#cCoRR
#index709
#id709
#%
#! analyze evaluate experimentally new probabilistic model word sequence prediction natural language based prediction suffix trees psts using efficient data structures extend notion pst unbounded vocabularies use bayesian approach based recursive priors possible psts efficiently maintain tree mixtures mixtures provably practically better performance single model evaluate model corpora low perplexity achieved relatively small pst mixture models suggests advantageous alternative theoretically practically widely used gram models

#* good turing smoothing tears
#@William Gale
#t1994
#cJournal of Quantitative Lin- guistics
#index710
#id710
#%
#! performance statistically based techniques tasks spelling correction sense disambiguation translation improved estimate probability object seen good turing methods means estimating probabilities previously unseen objects use good turing methods requires smoothing step smooth regions vastly different accuracy smoothers difficult use

#* model textual affect sensing using real world knowledge
#@Hugo Liu,Henry Lieberman,Ted Selker
#t2003
#cIUI
#index711
#id711
#%
#! paper presents novel way assessing affective qualities natural language scenario use previous approaches textual affect sensing employed keyword spotting lexical affinity statistical methods hand crafted models paper demonstrates new approach using large scale real world knowledge inherent affective nature everyday situations getting car accident classify sentences basic emotion categories

#* feature subset selection text learning
#@Dunja Mladenic
#t1998
#cECML
#index712
#id712
#%
#! paper describes known new methods feature subset selection large text data experimental comparison given real world data collected web users shows characteristics problem domain machine learning algorithm considered feature scoring measure selected problem domain consists hyperlinks given form small documents represented word vectors learning experiments naive bayesian classifier used

#* book recommending using text categorization extracted information
#@Raymond Mooney,Paul N. Bennett,Loriene Roy
#t1998
#cAAAI 1998 Workshop
#index713
#id713
#%
#! content based recommender systems suggest documents items services users based learning profile user rated examples containing information given items text categorization methods useful task generally rely unstructured text developed bookrecommending utilizes semi structured information items gathered web using simple information extraction techniques initial experimental results demonstrate

#* mining product reputions web
#@Satoshi Morinaga,Kenji Yamanishi,Kenji Tateishi,Toshikazu Fukushima
#t2002
#cKDD 2002
#index714
#id714
#%
#! knowing reputations competitors products important marketing customer relationship management costly collect analyze survey data manually paper presents new framework mining product reputations internet automatically collects people opinions target products web pages uses text mining techniques obtain reputations products advance generate basis human tested

#* learning surface text patterns question answering
#@Deepak Ravichandran,Eduard H. Hovy
#t2002
#cACL
#index715
#id715
#%
#! paper explore power surface text patterns open domain question answering systems order obtain optimal set patterns developed method learning patterns automatically tagged corpus built internet bootstrapping process providing hand crafted examples question type altavista patterns automatically extracted returned documents standardized calculate precision pattern average precision question type patterns applied answers new questions using trec question set report results cases answers determined trec corpus web

#* automatically generating extraction patterns untagged text
#@Ellen Riloff
#t1996
#cAAAI/IAAI, Vol. 2
#index716
#id716
#%
#! corpus based natural language processing systems rely text corpora manually annotated syntactic semantic tags particular previous dictionary construction systems information extraction used annotated training corpus form annotated input developed called autoslog ts creates dictionaries extraction patterns using untagged text autoslog ts based autoslog generated extraction

#* affect analysis text using fuzzy semantic typing
#@Pero Subasic,Alison Huettner
#t2000
#cIEEE Transactions on Fuzzy Systems
#index717
#id717
#%
#! propose novel convenient fusion natural language processing fuzzy logic techniques analyzing affect content free text main goals fast analysis visualization affect content decision making primary linguistic resource fuzzy semantic typing fuzzy affect lexicon important resources generated notably fuzzy thesaurus affect category groups free text tagged affect categories lexicon affect

#* phoaks sharing recommendations
#@Loren G. Terveen,William C. Hill,Brian Amento,David W. McDonald,Josh Creter
#t1997
#cCommun. ACM
#index718
#id718
#%
#! finding relevant high quality information theworld wide web difficult problem phoaks people helping know stuff experimental addresses problem collaborative filtering approach phoaks works automatically recognizing tallying redistributing recommendations web resources mined usenet news messages

#* unsupervised learning semantic orientation billion word corpus
#@Peter D. Turney,Michael L. Littman
#t2002
#cCoRR
#index719
#id719
#%
#! evaluative character word called semantic orientation positive semantic orientation implies desirability honest intrepid negative semantic orientation implies undesirability disturbing superfluous paper introduces simple algorithm unsupervised learning semantic orientation extremely large corpora method involves issuing queries web search engine using pointwise mutual information analyse results algorithm empirically evaluated using training corpus approximately billion words subset web indexed chosen search engine tested words positive negative algorithm attains accuracy test words include adjectives adverbs nouns verbs accuracy comparable results achieved hatzivassiloglou mckeown using complex stage supervised learning algorithm restricted determining semantic orientation adjectives

#* corpus study evaluative speculative language
#@Janyce Wiebe,Rebecca Bruce,Matthew Bell,Melanie Martin,Theresa Wilson
#t2001
#cAnnual Meeting of the ACL
#index720
#id720
#%
#! paper presents corpus study evaluative speculative language knowledge language useful applications text categorization summarization analyses annotator agreement characteristics subjective language performed study yields knowledge needed design effective machine learning systems identifying subjective language

#* identifying collocations recognizing opinions
#@Janyce Wiebe,Theresa Wilson,Matthew Bell
#t2001
#cProc. ACL-01 Workshop on Collocation: Computational Extraction, Analysis, and Exploitation
#index721
#id721
#%
#! subjectivity natural language refers aspects language used express opinions evaluations

#* using suffix arrays compute term frequency document frequency substrings corpus
#@Mikio Yamamoto,Kenneth Ward Church
#t2001
#cComputational Linguistics
#index722
#id722
#%
#! bigrams trigrams commonly used statistical natural language processing paper techniques working longer grams suffix arrays manber myers introduced compute frequency location substring gram sequence corpus length compute frequencies substrings corpus substrings grouped manageable number equivalence classes way prohibitive computation substrings reduced manageable computation classes paper presents algorithms code used compute term frequency tf document frequency dr grams large corpora english corpus million words wall street journal japanese corpus million characters mainichi shimbun second half paper uses frequencies interesting substrings lexicographers interested grams high mutual information mi joint term frequency higher expected chance assuming parts gram combine independently residual inverse document frequency ridf compares document frequency model chance terms particular term frequency distributed randomly collection mi tends pick phrases noncompositional semantics violate independence assumption ridf tends highlight technical terminology names good keywords information retrieval tend exhibit nonrandom distributions documents combination mi ridf better itself japanese word extraction task

#* opinion observer analyzing comparing opinions
#@Bing Liu,Minqing Hu,Junsheng Cheng
#t2005
#cInternational World Wide Web Conference
#index723
#id723
#%724,725,404,726,405,727,408,409,396,728,415,729,730,606,731,149,151,732,154,733
#! web excellent source gathering consumer opinions numerous web sites containing opinions customer reviews products forums discussion groups blogs paper focuses online customer reviews products makes contributions proposes novel framework analyzing comparing consumer opinions competing products prototype called opinion observer implemented single glance visualization user able clearly strengths weaknesses product minds consumers terms various product features comparison useful potential customers product manufacturers potential customer visual feature feature comparison consumer opinions products helps decide product buy product manufacturer comparison enables easily gather marketing intelligence product benchmarking information second new technique based language pattern mining proposed extract product features pros cons particular type reviews features form basis comparison experimental results technique highly effective outperform existing methods significantly

#* mining newsgroups using networks arising social behavior
#@Rakesh Agrawal,S. Rajagopalan,Ramakrishnan Srikant,Yirong Xu
#t2003
#cWWW
#index724
#id724
#%
#! recent advances information retrieval hyperlinked corpora convincingly demonstrated links carry noisy information text investigate feasibility applying link based methods new applications domains specific application consider partition authors opposite camps given topic context newsgroups typical newsgroup posting consists quoted lines posting followed opinion author social behavior gives rise network vertices individuals links represent responded relationships interesting characteristic newsgroups people frequently respond message disagree agree behavior sharp contrast www link graph linkage indicator agreement common analyzing graph structure responses able effectively classify people opposite camps contrast methods based statistical analysis text yield low accuracy datasets vocabulary used sides tends largely identical newsgroup postings consist relatively words text

#* collective information extraction relational markov networks
#@Razvan C. Bunescu,Raymond J. Mooney
#t2004
#cACL
#index725
#id725
#%
#! information extraction systems treat separate potential extractions independent cases considering influences different potential extractions improve overall accuracy statistical methods based undirected graphical models conditional random fields crfs shown effective approach learning accurate systems present new method employs relational markov networks generalization crfs represent arbitrary dependencies extractions allows collective information extraction exploits mutual influence possible extractions experiments learning extract protein names biomedical text demonstrate advantages approach

#* web scale information extraction knowitall preliminary results
#@Oren Etzioni,Michael Cafarella,Doug Downey,Stanley Kok,Ana-Maria Popescu,Tal Shaked,Stephen Soderland,Daniel S. Weld,Alexander Yates
#t2004
#cProceedings of the 13th international conference on World Wide Web
#index726
#id726
#%
#! manually querying search engines order accumulate large bodyof factual information tedious error prone process piecemealsearch search engines retrieve rank potentially relevantdocuments human perusal extract facts assessconfidence fuse information multiple documents paperintroduces knowitall aims automate tedious process ofextracting large collections facts web autonomous domain independent scalable manner paper describes preliminary experiments instance knowitall running days single machine able automatically extract facts knowitall associates probability fact enabling trade precision recall paper analyzes knowitall architecture reports lessons learned design large scale information extraction systems

#* information extraction hmm structures learned stochastic optimization
#@Dayne Freitag,A McCallum
#t2000
#cAAAI/IAAI
#index727
#id727
#%
#! recent research demonstrated strong performance hidden markov models applied information extraction task populating database slots corresponding phrases text documents remaining problem selection state transition structure model paper demonstrates extraction accuracy strongly depends selection structure presents algorithm automatically finding good structures stochastic optimization

#* conditional random fields probabilistic models segmenting labeling sequence data
#@J Lafferty,A McCallum,Fernando Pereira
#t2001
#cmachine learning-international workshop
#index728
#id728
#%
#! present conditional random fields framework building probabilistic models segment label sequence data conditional random fields offer advantages hidden markov models stochastic grammars tasks including ability relax strong independence assumptions models conditional random fields avoid fundamental limitation maximum entropy markov models memms discriminative markov models based directed graphical models biased states successor states present iterative parameter estimation algorithms conditional random fields compare performance resulting models hmms memms synthetic natural language data

#* mining data records web pages
#@
#t1000
#cEMPTY
#index729
#id729
#%
#! large information web contained regularly structured objects data records data records important present essential information host pages lists products services useful data records order extract information provide value added services existing automatic techniques satisfactory poor accuracies paper propose effective technique perform task technique based observations data records web string matching algorithm proposed technique able contiguous non contiguous data records experimental results proposed technique outperforms existing techniques substantially

#* hierarchical approach wrapper induction
#@Ion Muslea,Steven Minton,Craig A. Knoblock
#t1999
#cAgents
#index730
#id730
#%
#! tremendous information available web daily basis ability quickly develop information agents crucial problem vital component web based information agent set wrappers extract relevant data semistructured information sources novel approach wrapper induction based idea hierarchical information extraction turns hard problem extracting data arbitrarily complex document series easier extraction tasks introduce inductive algorithm stalker generates high accuracy extraction rules based user labeled training examples labeling training data represents major bottleneck using wrapper induction techniques experimental results stalker does significantly better approaches hand stalker requires orders magnitude fewer examples algorithms hand handle information sources wrapped existing techniques

#* robust metric opinion aaai spring symposium exploring attitude affect text
#@K Nigam,Matthew Hurst
#t2004
#cAAAI Spring Symposium on Exploring Attitude and Affect in Text
#index731
#id731
#%
#! paper describes automated detecting polar expressions topic elementary components approach shallow nlp polar language extraction machine learning based topic classifier components composed making simple accurate collocation assumption topical sentence contains polar language predicts polar language reflective topic subject matter evaluate components assumption corpus online consumer messages based components discuss measure overall sentiment particular topic expressed online messages authored different people propose use fundamentals bayesian statistics form aggregate authorial opinion metric metric propagate uncertainties introduced polarity topic modules facilitate statistically valid comparisons opinion multiple topics

#* classifying semantic relations bioscience text
#@Barbara Rosario,Marti A. Hearst
#t2004
#cAnnual Meeting of the Association for Computational Linguistics
#index732
#id732
#%
#! crucial step goal automatic extraction propositional information natural language text identification semantic relations constituents sentences examine problem distinguishing seven relation types occur entities treatment disease bioscience text problem identifying entities compare generative graphical models neural network using lexical syntactic semantic features finding help achieve high classification accuracy

#* web data extraction based partial tree alignment
#@Yanhong Zhai,Bing Liu
#t2005
#cWWW
#index733
#id733
#%
#! paper studies problem extracting data web page contains structured data records objective segment data records extract data items fields data database table problem studied researchers existing methods limitations class methods based machine learning requires human labeling examples web site interested extracting data process time consuming large number sites pages web second class algorithms based automatic pattern discovery methods inaccurate make assumptions paper proposes new method perform task automatically consists steps identifying individual data records page aligning extracting data items identified data records step propose method based visual information segment data records accurate existing methods step propose novel partial alignment technique based tree matching partial alignment means align data fields pair data records aligned matched certainty make commitment rest data fields approach enables accurate alignment multiple data records experimental results using large number web pages diverse domains proposed step technique able segment data records align extract data accurately

#* opinion retrieval blogs
#@Wei Zhang,Clement T. Yu,Weiyi Meng
#t2007
#cCIKM
#index734
#id734
#%404,735,408,396,598,736,146,737,738,149,148,150,739,740,741,742,743,744,745,746,747
#! opinion retrieval document retrieval process requires documents retrieved ranked according opinions query topic relevant document satisfy criteria relevant query topic contains opinions query matter positive negative paper opinion retrieval algorithm traditional information retrieval ir component topic relevant documents document set opinion classification component documents having opinions results ir step component rank documents based relevance query degrees having opinions query implemented algorithm working tested using trec blog track data automatic title runs result showed improvements map score best automatic runs track result higher state art opinion retrieval tested data set

#* determining semantic orientation terms gloss analysis
#@Andrea Esuli,Fabrizio Sebastiani
#t2005
#cProc. CIKM-05
#index735
#id735
#%
#! sentiment classification recent subdiscipline text classification concerned topic document opinion expresses rich set applications ranging tracking users opinions products political candidates expressed online forums customer relationship management functional extraction opinions text determination orientation subjective terms contained text determination term carries opinionated content positive negative connotation paper present new method determining orientation subjective terms method based quantitative analysis glosses terms definitions terms given line dictionaries use resulting term representations semi supervised term classification method present outperforms known methods tested recognized standard benchmarks task

#* major topic detection application opinion summarization
#@Lun-Wei Ku,Li-Ying Lee,Tung-Ho Wu,Hsin-Hsi Chen
#t2005
#cSIGIR
#index736
#id736
#%
#! watching specific information sources summarizing newly discovered opinions important governments improve services companies improve products queries posed detecting opinions similar task topic detection sentence level telling opinions positive negative identifying events correlated opinions important paper proposes major topic detection mechanism capture main concepts embedded implicitly relevant document set opinion summarization retrieves relevant sentences related major topic document set determines opinion polarity relevant sentence finally summarizes positive negative sentences respectively

#* effective approach document retrieval utilizing wordnet recognizing phrases
#@Shuang Liu,Fang Liu,Clement T. Yu,Weiyi Meng
#t2004
#cSIGIR
#index737
#id737
#%
#! noun phrases queries identified classified types proper names dictionary phrases simple phrases complex phrases document phrase content words phrase window certain size window sizes different types phrases different determined using decision tree phrases important individual terms consequently documents response query ranked matching phrases given higher priority utilize wordnet disambiguate word senses query terms sense query term determined synonyms hyponyms words definition compound words considered possible additions query experimental results approach yields improvements best known results trec collections short title queries using web data

#* overview trec blog track
#@Iadh Ounis,Maarten de Rijke,Craig Macdonald,G Mishne,Ian Soboroff
#t2006
#cTREC 2006
#index738
#id738
#%
#! rise internet blogging creation journal like web page logs created highly dynamic subset world wide web evolves responds real world events blogs weblogs recently emerged new grassroots publishing medium called blogosphere collection blogs internet opens new interesting research areas blogs interesting features entries added chronological order high volume addition blogs created authors intended sizable audience purely mechanism self expression extremely accessible blog software facilitated act blogging wide ranging audience blogs reflecting opinions philosophies emotions traditional media tends focus heavy hitting blogs devoted politics punditry technology different genres blogs written specific topic covering talking personal daily life blog track began year aim explore information seeking behaviour blogosphere purpose new large scale test collection trec blog collection created pilot run track tasks main task opinion retrieval open task opinion retrieval task focuses specific aspect blogs opinionated nature blogs second task introduced allow participants opportunity influence determination suitable second task aspects blogs temporal event related nature blogs severity spam blogosphere

#* using appraisal groups sentiment analysis
#@Casey Whitelaw,Navendu Garg,Shlomo Argamon
#t2005
#cCIKM
#index739
#id739
#%748,601,749,750,404,408,751,752,606,195,149,753,754,755
#! little work date sentiment analysis classifying texts positive negative orientation attempted use fine grained semantic distinctions features used classification present new method sentiment classification based extracting analyzing appraisal groups good terribly funny appraisal group represented set attribute values task independent semantic taxonomies based appraisal theory semi automated methods used build lexicon appraising adjectives modifiers classify movie reviews using features based taxonomies combined standard bag words features report state art accuracy addition types appraisal appear significant sentiment classification

#* learning subjective language
#@Janyce Wiebe,Theresa Wilson,Rebecca Bruce,Matthew Bell,Melanie Martin
#t2004
#cComputational Linguistics
#index740
#id740
#%
#! subjectivity natural language refers aspects language used express opinions evaluations speculations numerous natural language processing applications subjectivity analysis relevant including information extraction text categorization goal work learning subjective language corpora clues subjectivity generated tested including low frequency words collocations adjectives verbs identified using distributional similarity features examined working concert features generated different data sets using different procedures exhibit consistency performance better worse data sets addition article shows density subjectivity clues surrounding context strongly affects likely word subjective provides results annotation study assessing subjectivity sentences high density features finally clues used perform opinion piece recognition type text categorization genre detection demonstrate utility knowledge acquired article

#* comparative study feature selection text categorization
#@Yiming Yang,Jan O. Pedersen
#t1997
#cICML
#index741
#id741
#%
#! paper comparative study feature selection methods statistical learning text categorization focus aggressive dimensionality reduction methods evaluated including term selection based document frequency df information gain ig mutual information mi test chi term strength ts ig chi effective experiments using ig thresholding knearest neighbor classifier reuters corpus removal

#* multiple ranking strategies opinion retrieval blogs
#@G Mishne
#t0
#cnull
#index742
#id742
#%
#! participation opinion retrieval task trec approach identifying opinions blog post consisted scoring posts separately various aspects associated expression opinion topic including shallow sentiment analysis spam detection link based authority estimation separate approaches combined single ranking yielding significant improvement content baseline

#* using blog properties improve retrieval
#@G Mishne
#t2007
#cProceedings of ICWSM
#index743
#id743
#%
#! paper describes simple heuristics improve opinion retrieval effectiveness using blog specific properties blog timestamps used increase retrieval scores blog posts published near time significant event related query inexpensive approach comment estimation used identify level opinion expressed post query specific weights used change importance spam filtering different types queries overall methods combined non blogspecific retrieval approaches result substantial improvements state art

#* widit trec blog track
#@Kiduk Yang,Ning Yu,Alejandro Valerio,Hui Zhang
#t2006
#cTREC
#index744
#id744
#%
#! web information discovery integrated tool widit laboratory indiana university school library information science participated blog tracks opinion task trec goal opinion task uncover public sentiment given entity target involves retrieving topically relevant blogs identifying contain opinions target complicate matter blog test collection contains considerable noise blogs non english content non blog content advertisement navigational text misdirect retrieval systems based hypothesis noise reduction exclusion non english blogs navigational text improve topic opinion retrieval performances explored various noise reduction approaches effectively eliminate noise blog data inadvertently excluding valid content creating separate indexes noise assess noise reduction effect tackled opinion blog retrieval task breaking sequential subtasks topic retrieval followed opinion classification opinion retrieval approach apply traditional ir methods retrieve topic blogs boost ranks opinionated blogs based opinion scores generated opinion assessment methods opinion module consists opinion term module identify opinions based frequency opinion terms terms occur frequently opinion blogs rare term module uses uncommon rare terms sooo good opinion classification iu module uses iu collocations adjective verb module uses computational linguistics distribution similarity approach learn subjective language training data

#* trec maryland blog enterprise legal qa tracks
#@Douglas Oard,Tamer Elsayed,Jianqiang Wang,Yejun Wu,Pengyi Zhang,Eileen Abels,Jimmy Lin,Dagbert Soergel
#t2006
#cTREC
#index745
#id745
#%
#! trec teams university maryland participated blog track expert search task enterprise track complex interactive question answering task question answering track legal track paper reports results

#* recognition classification noun phrases queries effective retrieval
#@Wei Zhang,Shuang Liu,Clement T. Yu,Chaojing Sun,Fang Liu,Weiyi Meng
#t2007
#cCIKM
#index746
#id746
#%
#! shown using phrases properly document retrieval leads higher retrieval effectiveness paper define types noun phrases present algorithm recognizing phrases queries strengths existing tools combined phrase recognition algorithm tested using set web queries query log set trec queries experimental results algorithm yields high phrase recognition accuracy use baseline noun phrase recognition algorithm recognize phrases trec queries document retrieval experiment conducted using trec queries phrases phrases recognized baseline noun phrase recognition algorithm phrases recognized algorithm respectively retrieval effectiveness better better demonstrates utilizing phrases queries does improve retrieval effectiveness better noun phrase recognition yields higher retrieval performance

#* query expansion using local global document analysis
#@Jinxi Xu,W. Bruce Croft
#t1996
#cSIGIR
#index747
#id747
#%
#! automatic query expansion long suggested technique dealing fundamental issue word mismatch information retrieval number approaches ezpanrnion studied recently attention focused techniques analyze corpus discover word relationship global techniques analyze documents retrieved initial quer local feedback paper compare effectiveness approaches global analysis haa advantages local analysia generally effective using global analysis techniques word contezt phrase structure local aet documents produces results effective predictable simple local feedback

#* conjunction modal assessment genre classification
#@Shlomo Argamon,Jeff Dodick
#t2004
#cAAAI Spring Symposium on Attitude and Affect in Text
#index748
#id748
#%
#! use textual features motivated systemic functional linguistic theory genre based text categorization developed feature sets representing different types conjunctions modal assessment indicate partially different genres structure texts express attitudes propositions text using features enables analysis large scale rhetorical differences genres examining features important classification specific domain studied comprises scientific articles historical experimental sciences paleontology physical chemistry respectively smo learning algorithm feature set achieved accuracy classifying articles according field field specific terms used features highly weighted features consistent hypothesized methodological differences historical experimental sciences lending empirical evidence notion multiple scientific methods

#* introduction support vector machines
#@Nello Cristianini,John Shawe-Taylor
#t2000
#cbook
#index749
#id749
#%
#! comprehensive introduction support vector machines svms new generation learning based recent advances statistical learning theory svms deliver state art performance real world applications text categorisation hand written character recognition image classification biosequences analysis established standard tools machine learning data mining students book stimulating accessible practitioners guided smoothly material required good grasp theory applications concepts introduced gradually accessible self contained stages presentation rigorous thorough pointers relevant literature web sites containing software ensure forms ideal starting point study equally book associated web site guide practitioners updated literature new applications line software

#* ensemble methods automatic thesaurus extraction
#@James R. Curran
#t2002
#cProc. Conference on Empirical Methods in Natural Language Processing
#index750
#id750
#%
#! ensemble methods state art nlp tasks recent work banko brill suggests necessarily true large training corpora available results limited simplicity evaluation task individual classifiers work explores ensemble efficacy complex task automatic thesaurus extraction million words examine conflicting results terms constraints complexity different contextual representations contribute sparseness noise induced bias behaviour nlp systems large corpora

#* rhetorical parsing unrestricted texts surface based approach
#@Daniel Marcu
#t2000
#cComputational Linguistics
#index751
#id751
#%
#! coherent texts just simple sequences clauses sentences complex artifacts highly elaborate rhetorical structure paper explores extent formed rhetorical structures automatically derived means surface form based algorithms algorithms identify discourse usages cue phrases break sentences clauses hypothesize rhetorical relations hold textual units produce valid rhetorical structure trees unrestricted natural language texts algorithms empirically grounded corpus analysis cue phrases rely order formalization rhetorical structure trees algorithms evaluated intrinsically extrinsically intrinsic evaluation assesses resemblance automatically manually constructed rhetorical structure trees extrinsic evaluation shows automatically derived rhetorical structures successfully exploited context text summarization

#* sentiment analysis using support vector machines diverse information sources
#@Mullen Tony,Collier Nigel
#t2004
#cProceedings of EMNLP-04, 9th Conference on Empirical Methods in Natural Language Processing
#index752
#id752
#%
#! paper introduces approach sentiment analysis uses support vector machines svms bring diverse sources potentially pertinent information including favorability measures phrases adjectives available knowledge topic text models using features introduced combined unigram models shown effective past pang et al lemmatized versions unigram models experiments movie review data internet movie database demonstrate hybrid svms combine unigram style feature based svms based real valued favorability measures obtain superior performance producing best results published using data experiments using feature set enriched topic information smaller dataset music reviews hand annotated topic reported results suggest incorporating topic information models yield improvement

#* fast training support vector machines using sequential minimal optimization
#@John C. Platt
#t1999
#cAdvances in kernel methods: support vector learning (1999)
#index753
#id753
#%
#! chapter describes new algorithm training support vector machines sequential minimal optimization smo training support vector machine svm requires solution large quadratic programming optimization problem smo breaks large problem series smallest possible problems small problems solved analytically avoids using time consuming numerical optimization inner loop memory required smo linear training set size allows smo handle large training sets large matrix computation avoided smo scales linear quadratic training set size various test problems standard projected conjugate gradient pcg chunking algorithm scales linear cubic training set size smo computation time dominated svm evaluation smo fastest linear svms sparse data sets mnist database smo fast pcg chunking uci adult database linear svms smo times faster pcg chunking algorithm

#* analyzing appraisal automatically
#@Maite Taboada,Jack Grieve
#t2004
#cAAAI Spring Symposium on Exploring Attitude and Affect in Text: Theories and Applications
#index754
#id754
#%
#! present method classifying texts automatically based subjective content apply stan dard method calculating semantic orientation tur ney expand giving prominence tocertain parts text believe subjec tive content concentrated apply linguisticclassification appraisal help ful distinguishing different types subjective texts movie reviews consumer product reviews

#* identifying interpersonal distance using systemic features
#@Maria Herke-Couchman,Casey Whitelaw,Jon Patrick
#t2004
#cAAAI
#index755
#id755
#%
#! chapter uses systemic functional linguistic sfl theory basis extracting semantic features documents focus pronominal determination role plays constructing interpersonal distance using hierarchical model represents authors language choices possible construct richer informative feature representation superior computational efficiency usual bag words approach experiments context financial scam classification systemic features create clear separation registers different interpersonal distance approach generalizable aspects attitude affect modelled systemic functional linguistic theory

#* opinion spam analysis
#@Nitin Jindal,Bing Liu
#t2008
#cWSDM
#index756
#id756
#%658,404,757,396,758,149,150,759,760,761,762
#! evaluative texts web valuable source opinions products services events individuals recently researchers studied opinion sources product reviews forum posts blogs existing research focused classification summarization opinions using natural language processing data mining techniques important issue neglected far opinion spam trustworthiness online opinions paper study issue context product reviews opinion rich widely used consumers product manufacturers past years startup companies appeared aggregate opinions product reviews high time study spam reviews best knowledge published study topic web spam email spam investigated extensively opinion spam quite different web spam email spam requires different detection techniques based analysis million reviews million reviewers amazon com opinion spam reviews widespread paper analyzes spam activities presents novel techniques detect

#* web spam taxonomy
#@Zoltan Gyongyi,Hector Garcia-Molina
#t2005
#cAIRWeb
#index757
#id757
#%
#! web spamming refers actions intended mislead search engines ranking pages higher deserve recently web spam increased dramatically leading degradation search results paper presents comprehensive taxonomy current spamming techniques believe help developing appropriate countermeasures

#* detecting spam web pages content analysis
#@Alexandros Ntoulas,Marc Najork,Mark Manasse,Dennis Fetterly
#t2006
#cWWW
#index758
#id758
#%
#! paper continue investigations web spam injection artificially created pages web order influence results search engines drive traffic certain pages fun profit paper considers previously undescribed techniques automatically detecting spam pages examines effectiveness techniques isolation aggregated using classification algorithms combined heuristics correctly identify spam pages judged collection pages misidentifying spam non spam pages

#* spam double funnel connecting web spammers advertisers
#@Yi-Min Wang,Ming Ma,Yuan Niu,Hao Chen
#t2007
#cWWW
#index759
#id759
#%
#! spammers use questionable search engine optimization seo techniques promote spam links search results paper focus prevalent type spam redirection spam identify spam pages party domains pages redirect traffic propose fivelayer double funnel model describing end end redirection spam present methodology analyzing layers identify prominent domains layer using sets commercial keywords targeting spammers targeting advertisers methodology findings useful search engines strengthen ranking algorithms spam legitimate website owners locate remove spam doorway pages legitimate advertisers identify unscrupulous syndicators serve ads spam pages

#* identifying link farm spam pages
#@Baoning Wu,Brian D. Davison
#t2005
#cWWW (Special interest tracks and posters)
#index760
#id760
#%
#! increasing importance search guiding today web traffic effort spent create search engine spam link analysis important factors current commercial search engines ranking systems new kinds spam aiming links appeared building link farms technique deteriorate link based ranking algorithms paper present algorithms detecting link farms automatically generating

#* topical trustrank using topicality combat web spam
#@Baoning Wu,Vinay Goel,Brian D. Davison
#t2006
#cWWW
#index761
#id761
#%
#! web spam behavior attempts deceive search engine ranking algorithms trustrank recent algorithm combat web spam trustrank vulnerable sense seed set used trustrank sufficiently representative cover different topics web given seed set trustrank bias larger communities propose use topical information partition seed set calculate trust scores topic separately address issues combination trust scores page used determine ranking experimental results large datasets topical trustrank better performance trustrank demoting spam sites pages compared trustrank best technique decrease spam ranked sites

#* utility scoring product reviews
#@Zhu Zhang,Balaji Varadarajan
#t2006
#cCIKM
#index762
#id762
#%763,404,408,614,146,195,148,149,150,151,764,765,85,766,767,420,740,768,154
#! identify new task ongoing research text sentiment analysis predicting utility product reviews orthogonal polarity classification opinion extraction build regression models incorporating diverse set features achieve highly competitive performance utility scoring real world data sets

#* identifying sources opinions conditional random fields extraction patterns
#@Yejin Choi,Claire Cardie,Ellen Riloff,Siddharth Patwardhan
#t2005
#cHLT/EMNLP
#index763
#id763
#%
#! recent systems developed sentiment classification opinion recognition opinion analysis detecting polarity strength pursue aspect opinion analysis identifying sources opinions emotions sentiments view problem information extraction task adopt hybrid approach combines conditional random fields lafferty et al variation autoslog riloff crfs model source identification sequence tagging task autoslog learns extraction patterns results combination methods performs better resulting identifies opinion sources precision recall using head noun matching measure precision recall using overlap measure

#* exploiting subjectivity classification improve information extraction
#@Ellen Riloff,Janyce Wiebe,William Phillips
#t2005
#cAAAI
#index764
#id764
#%
#! information extraction systems prone false hits variety reasons observed false hits occur sentences contain subjective language opinions emotions sentiments motivated observations explore idea using subjectivity analysis improve precision information extraction systems paper uses subjective sentence classifier filter extractions experimented different strategies using subjectivity classifications including aggressive strategy discards extractions subjective sentences complex strategies selectively discard extractions evaluated performance different approaches muc terrorism data set indiscriminately filtering extractions subjective sentences overly aggressive selective filtering strategies improved precision minimal recall loss introduction

#* learning subjective nouns using extraction pattern bootstrapping
#@Ellen Riloff,Janyce Wiebe,Theresa Wilson
#t2003
#c2003	Proceeings of the Seventh CoNLL conference
#index765
#id765
#%
#! explore idea creating subjectivity classifier uses lists subjective nouns learned bootstrapping algorithms goal research develop distinguish subjective sentences objective sentences use bootstrapping algorithms exploit extraction patterns learn sets subjective nouns train naive bayes classifier using subjective nouns discourse features subjectivity clues identified prior

#* multi perspective question answering using opqa corpus
#@Veselin Stoyanov,Claire Cardie,Janyce Wiebe
#t2005
#cHLT/EMNLP
#index766
#id766
#%
#! investigate techniques support answering opinion based questions present opqa corpus opinion questions answers using corpus compare contrast properties fact opinion questions answers based disparate characteristics opinion vs fact answers argue traditional fact based qa approaches difficulty mpqa setting modification initial step development mpqa systems investigate use machine learning rule based subjectivity opinion source filters used guide mpqa systems

#* bootstrapping method learning semantic lexicons using extraction pattern contexts
#@Michael Thelen,Ellen Riloff
#t2002
#cEMNLP
#index767
#id767
#%
#! paper describes bootstrapping algorithm called basilisk learns high quality semantic lexicons multiple categories basilisk begins unannotated corpus seed words semantic category bootstrapped learn new words category basilisk hypothesizes semantic class word based collective information large body extraction pattern contexts evaluate basilisk semantic categories semantic lexicons produced basilisk higher precision produced previous techniques categories showing substantial improvement

#* recognizing contextual polarity phrase level sentiment analysis
#@Theresa Wilson,Janyce Wiebe,Paul Hoffmann
#t2005
#cHLT/EMNLP
#index768
#id768
#%
#! paper presents new approach phrase level sentiment analysis determines expression neutral polar disambiguates polarity polar expressions approach able automatically identify contextual polarity large subset sentiment expressions achieving results significantly better baseline

#* review spam detection
#@Nitin Jindal,Bing Liu
#t2007
#cWWW
#index769
#id769
#%658,757,396,770,758,761
#! common practice commerce web sites enable customers write reviews products purchased reviews provide valuable sources information products used potential customers opinions existing users deciding purchase product used product manufacturers identify problems products competitive intelligence information competitors unfortunately importance reviews gives good incentive spam contains false positive malicious negative opinions paper make attempt study review spam spam detection best knowledge reported study problem

#* fast statistical spam filter approximate classifications
#@Kang Li,Zhenyu Zhong
#t2006
#cSIGMETRICS/Performance
#index770
#id770
#%
#! statistical based bayesian filters popular important defense spam despite effectiveness greater processing overhead prevent scaling enterprise level mail servers example dictionary lookups characteristic approach limited memory access rate relatively insensitive increases cpu speed address scaling issue proposing acceleration technique speeds bayesian filters based approximate classification approximation uses methods hash based lookup lossy encoding lookup approximation based popular bloom filter data structure extension support value retrieval lossy encoding used compress data structure methods introduce additional errors strict bayesian approach errors minimized biased false negative classification demonstrate speedup known spam filters bogofilter qsf achieving identical false positive rate similar false negative rate original filters

#* sentiment analysis capturing favorability using natural language processing
#@Nasukawa Tetsuya,Yi Jeonghee
#t2003
#c K-CAP '03
#index771
#id771
#%149,772,408,369,773,721
#! paper illustrates sentiment analysis approach extract sentiments associated polarities positive negative specific subjects document instead classifying document positive negative essential issues sentiment analysis identify sentiments expressed texts expressions indicate positive favorable negative unfavorable opinions subject order improve accuracy sentiment analysis important properly identify semantic relationships sentiment expressions subject applying semantic analysis syntactic parser sentiment lexicon prototype achieved high precision depending data finding sentiments web pages news articles

#* message understanding conference brief history
#@Ralph Grishman,Beth Sundheim
#t1996
#c1996	International Conference On Computational Linguistics
#index772
#id772
#%
#! recently completed sixth series message understanding conferences designed promote evaluate research information extraction muc introduced innovations prior mucs notably range different tasks evaluations conducted motivations new format briefly discuss results evaluations

#* talent textract architecture data model
#@Mary S. Neff,Roy J. Byrd,Branimir K. Boguraev
#t2004
#cNatural Language Engineering
#index773
#id773
#%
#! present architecture data model textract robust scalable configurable document analysis framework textract engineered pipeline architecture allowing rapid prototyping application development freely mixing reusable existing language analysis plugins custom new plugins customizable functionality discuss design issues arise requirements industrial strength efficiency scalability constrained plugin interactions common data model comprising annotation store document vocabulary lexical cache exemplify focusing meta plugin interpreter annotation based finite state transduction linguistic filters implemented stand plugins framework component plugins extensively deployed research industrial environments broad range text analysis mining tasks

#* sentiment analyzer extractingsentiments given topic
#@Jeonghee Yi,Tetsuya Nasukawa,Razvan Bunescu,Wayne Niblack
#t2003
#cIEEE Intl. Conf. on Data Mining (ICDM)
#index774
#id774
#%775,404,776,409,777,778,369,779,780,149,781,782,717,718,420,783,784
#! present sentiment analyzer sa extracts sentiment opinion subject online text documents instead classifying sentiment entire document subject sa detects references given subject determines sentiment references using natural language processing nlp techniques sentiment analysis consists topic specific feature term extraction sentiment extraction subject sentiment association relationship analysis sa utilizes linguistic resources analysis sentiment lexicon sentiment pattern database performance algorithms verified online product review articles digital camera music reviews general documents including general webpages news articles

#* finding parts large corpora
#@Matthew Berland,Eugene Charniak
#t1999
#cACL
#index775
#id775
#%
#! present method extracting parts objects wholes speedometer car given large corpus method finds words accuracy words ranked list scanned end user added existing ontology wordnet used rough semantic lexicon

#* accurate methods statistics surprise coincidence
#@Ted Dunning
#t1993
#cComputational Linguistics
#index776
#id776
#%
#! work statistical analysis text cases reported literature inappropriate statistical methods used statistical significance results addressed particular asymptotic normality assumptions used unjustifiably leading flawed results assumption normal distribution limits ability analyze rare events unfortunately rare events make large fraction real text applicable methods based likelihood ratio tests available yield good results relatively small samples tests implemented efficiently used detection composite terms determination domain specific terms cases measures perform better methods previously used cases traditional contingency table methods work likelihood ratio tests described nearly identical paper describes basis measure based likelihood ratios applied analysis text

#* sentence processing information access world wide web
#@Boris Katz
#t1997
#cProc. AAAI Spring Symposium on Natural Language Processing for the World Wide Web
#index777
#id777
#%
#! paper describes start information server built mit artificial intelligence laboratory available world wide web december start server provides users access multi media information response questions formulated english years start server answered millions questions users world start server built foundations sentence level natural language processing capability provided start natural language katz idea natural language annotations multi media information segments paper starts overview sentence level processing start explains annotating information segments collections english sentences makes possible use power sentence level natural language processing service multi media information access paper ends proposal annotate world wide web

#* mining open answers questionnaire data
#@Hang Li,Kenji Yamanishi
#t2001
#cKDD
#index778
#id778
#%
#! surveys important tools marketing managing customer relationships answers open ended questions particular contain valuable information provide important basis business decisions summaries human analysts make open answers tend rely intuition

#* marcinkiewicz building large annotated corpus english penn treebank
#@Mitchell P. Marcus,Mary Ann Marcinkiewicz,Beatrice Santorini
#t1994
#cComputational Linguistics
#index779
#id779
#%
#! paper review experience constructing large annotated corpus penn treebank corpus

#* nouns wordnet lexical inheritance
#@George A. Miller
#t1990
#cInt J Lexicography
#index780
#id780
#%
#! prototypical definition noun consists immediate superordinate followed relative clause describes instance differs instances information provides basis organizing files wordnet wordnet lexical inheritance hyponyms connected superordinates vice versa lexical database hierarchy searched upward downward equal speed

#* maximum entropy model ofspeech tagging
#@Adwait Ratnaparkhi
#t1996
#cEmpirical Methods in Natural Language Processing
#index781
#id781
#%
#! paper presents statistical model trains corpus annotated ofspeech tags assigns previously unseen text state art accuracy model classified maximum entropy model simultaneously uses contextual features predict pos tag furthermore paper demonstrates use specialized features model difficult tagging decisions discusses corpus consistency problems discovered implementation

#* emotion style second television advertisements targeted men women boys girls
#@L. Rovinelli, C. Whissell
#t1998
#c Percept Mot Skills
#index782
#id782
#%
#! program objective textual analysis incorporated measures style word emotionality word imagery used score verbal portion sec television advertisements analysis indicated advertisements directed children active longer negative directed adults comparison advertisements directed males females regardless age showed greater linguistic complexity words fewer common words text directed women girls stylistic emotional measures used advertisements produced significant difference associated age sex target population interaction

#* model based feedback lanuage modeling approach information retrieval
#@ChengXiang Zhai,J Lafferty
#t2001
#c2001	Proceedings of Tenth International Conference on Information and Knowledge Management
#index783
#id783
#%
#! language modeling approach retrieval shown perform empirically advantage new approach statistical foundations feedback important component retrieval dealt heuristically new retrieval approach original query usually literally expanded adding additional terms expansion based feedback creates inconsistent interpretation original expanded query paper present principled approach feedback language modeling approach specifically treat feedback updating query language model based extra evidence carried feedback documents model based feedback strategy easily fits extension language modeling approach propose evaluate different approaches updating query language model based feedback documents based generative probabilistic model feedback documents based minimization kl divergence feedback documents experiment results approaches effective outperform rocchio feedback approach

#* exact maximum likelihood estimation word mixtures
#@Yi Zhang,Wei Xu,Jamie Callan
#t2002
#cICML
#index784
#id784
#%
#! mixture model generating document generative language model used information retrieval using model situations need maximum likelihood estimation density multinomial given fixed mixture weight densities multinomial

#* topic sentiment mixture
#@Qiaozhu Mei,Xu Ling,Matthew Wondra,Hang Su,ChengXiang Zhai
#t2007
#cWWW 2007
#index785
#id785
#%182,763,185,186,786,205,787,190,788,146,193,789,194,790,791,195,148,149,328,792,793,794,608
#! paper define problem topic sentiment analysis weblogs propose novel probabilistic model capture mixture topics sentiments simultaneously proposed topic sentiment mixture tsm model reveal latent topical facets weblog collection subtopics results ad hoc query associated sentiments provide general sentiment models applicable ad hoc topics specifically designed hmm structure sentiment models topic models estimated tsm utilized extract topic life cycles sentiment dynamics empirical experiments different weblog datasets approach effective modeling topic facets sentiments extracting dynamics weblog collections tsm model quite general applied text collections mixture topics sentiments potential applications search result summarization opinion tracking user behavior prediction

#* clustering versus faceted categories information exploration
#@Marti A. Hearst
#t2006
#cCommun. ACM
#index786
#id786
#%
#! information seekers express desire user interface organizes search results meaningful groups order help make sense results help decide longitudinal study participants provided ability group search results changed search habits response having grouping mechanism available open research questions generate useful groupings design interfaces support exploration using grouping currently methods quite popular clustering faceted categorization approaches summarize advantages disadvantages based results usability studies

#* viewing morphology inference process
#@Robert Krovetz
#t2000
#cArtif. Intell.
#index787
#id787
#%
#! morphology area linguistics concerned internal structure words information retrieval generally paid attention word structure account variability word forms use stemmers paper experiments determine importance morphology effect performance role morphological analysis word sense disambiguation identifying lexical

#* pachinko allocation dag structured mixture models topic correlations
#@Wei Li,A McCallum
#t2006
#cICML
#index788
#id788
#%
#! latent dirichlet allocation lda related topic models increasingly popular tools summarization manifold discovery discrete data lda does capture correlations topics paper introduce pachinko allocation model pam captures arbitrary nested possibly sparse correlations topics using directed acyclic graph dag leaves dag represent individual words vocabulary interior node represents correlation children words interior nodes topics pam provides exible alternative recent work blei la erty captures correlations pairs topics using text data newsgroups historic nips proceedings research paper corpora improved performance pam document classification likelihood held data ability support finer grained topics topical keyword coherence

#* discovering evolutionary theme patterns text exploration temporal text mining
#@Qiaozhu Mei,ChengXiang Zhai
#t2005
#cKDD
#index789
#id789
#%
#! temporal text mining ttm concerned discovering temporal patterns text information collected time text information bears time stamps ttm applications multiple domains summarizing events news articles revealing research trends scientific literature paper study particular ttm task discovering summarizing evolutionary patterns themes text stream define new text mining problem present general probabilistic methods solving problem discovering latent themes text constructing evolution graph themes analyzing life cycles themes evaluation proposed methods different domains news articles literature shows proposed methods discover interesting evolutionary theme patterns effectively

#* moodviews tools blog mood analysis
#@G Mishne,M de Rijke
#t2006
#cAAAI
#index790
#id790
#%
#! demonstrate tracking analyzing moods bloggers worldwide reflected largest blogging community livejournal collects thousands blog posts hour performs various analyses posts presents results graphically

#* predicting movie sales blogger sentiment
#@G Mishne,N Glance
#t2006
#cAAAI
#index791
#id791
#%
#! volume discussion product weblogs recently shown correlate products financial performance paper study applying sentiment analysis methods weblog data results better correlation volume domain movies main finding positive sentiment better predictor movie success applied limited context references movie weblogs posted prior release

#* regularized estimation mixture models robust pseudo relevance feedback
#@Tao Tao,ChengXiang Zhai
#t2006
#cSIGIR
#index792
#id792
#%
#! pseudo relevance feedback proven effective strategy improving retrieval accuracy retrieval models performance existing pseudo feedback methods affected significantly parameters number feedback documents use relative weight original query terms parameters generally set trial error guidance paper present robust method pseudo feedback based statistical language models main idea integrate original query feedback documents single probabilistic mixture model regularize estimation language model parameters model information feedback documents gradually added original query unlike existing feedback methods new method parameter tune experiment results representative data sets new method significantly robust state art baseline language modeling approach feedback comparable better retrieval accuracy

#* annotating expressions opinions emotions language
#@Janyce Wiebe,Theresa Wilson,Claire Cardie
#t2005
#cLanguage Resources and Evaluation
#index793
#id793
#%
#! paper describes corpus annotation project study issues manual annotation opinions emotions sentiments speculations evaluations private states language resulting corpus annotation scheme described examples use addition manual annotation process results inter annotator agreement study sentence corpus articles drawn world press presented

#* sentiment analyzer extracting sentiments given topic using natural language processing techniques
#@Jeonghee Yi,Tetsuya Nasukawa,Razvan C. Bunescu,Wayne Niblack
#t2003
#cICDM
#index794
#id794
#%
#! present sentiment analyzer sa extracts sentiment opinion subject online text documents instead classifying sentiment entire document subject sa detects references given subject determines sentiment references using natural language processing nlp techniques sentiment analysis consists topic specific feature term extraction sentiment extraction subject sentiment association relationship analysis sa utilizes linguistic resources analysis sentiment lexicon sentiment pattern database performance algorithms verified online product review articles digital camera music reviews general documents including general webpages news articles

#* using polarity scores words sentence level opinion extraction
#@Lun-Wei Ku,Yong-Sheng Lo ,Hsin-Hsi Chen
#t2007
#c2007	Proceedings of NTCIR-6 Workshop Meeting
#index795
#id795
#%401,404,396,143,736,149,151,796,797,798,799,765,145
#! opinion analysis task pilot study task ntcir contains challenges opinion sentence extraction opinion polarity judgment opinion holder extraction relevance sentence extraction new tasks proven tough trec paper introduce analyzing opinionated information formulae proposed decide opinion polarities strengths words composed characters process opinion sentences negation operators taken consideration opinion polarity judgment opinion operators used clues locations opinion holders performance opinion extraction polarity judgment achieves measure lenient metric strict metric second best participants

#* extracting semantic orientations words using spin model
#@Hiroya Takamura,Takashi Inui,Manabu Okumura
#t2005
#cACL
#index796
#id796
#%
#! propose method extracting semantic orientations words desirable undesirable regarding semantic orientations spins electrons use mean field approximation compute approximate probability function instead intractable actual probability function propose criterion parameter selection basis magnetization given small number seed words proposed method extracts semantic orientations high accuracy experiments english lexicon result comparable best value reported

#* identify collocations recognizing opinions
#@Janyce Wiebe,Theresa Wilson,Matthew Bell
#t2001
#cProc. ACL-01 Workshop on Collocation: Computational Extraction, Analysis, and Exploitation
#index797
#id797
#%
#! subjectivity natural language refers aspects language used express opinions evaluations

#* overview trec novelty track
#@Ian Soboroff,Donna Harman
#t2003
#cTREC
#index798
#id798
#%
#! novelty track introduced trec given trec topic ordered list documents systems relevant novel sentences returned user set task integrates aspects passage retrieval information filtering year using old trec topics documents developed fifty new topics specifically novelty track topics classes events opinions additionally documents

#* learning parsimonious models extracting consumer opinions
#@Xue Bai,Rema Padman,Edoardo Airoldi
#t2005
#cHICSS
#index799
#id799
#%
#! extracting sentiments unstructured text emerged important problem disciplines accurate method enable example online opinions internet learn customers preferences economic marketing research leveraging strategic advantage paper propose stage bayesian algorithm able capture dependencies words time finds vocabulary efficient purpose extracting sentiments experimental results movie reviews data set algorithm able select parsimonious feature set substantially fewer predictor variables data set leads better predictions sentiment orientations state art machine learning methods findings suggest sentiments captured conditional dependence relations words keywords high frequency words

#* clustering methods collaborative filtering
#@Lyle H. Ungar,Dean Foster,Ellen Andre,Star Wars,Fred Star Wars,Dean Star Wars,Jason Hiver Whispers
#t1998
#cWorkshop on Recommendation Systems
#index800
#id800
#%801,802
#! grouping people clusters based items purchased allows accurate recommendations new items purchase liked movies probably enjoy movies like recommending items based similarity collaborative filtering attractive domains books cds movies does work data sparse given person seen small fraction movies accurate predictions grouping people clusters similar movies grouping movies clusters tend liked people finding optimal clusters tricky movie groups used help determine people groups visa versa present formal statistical model collaborative filtering compare different algorithms estimating model parameters including variations means clustering gibbs sampling formal model easily extended handle clustering objects multiple attributes

#* using collaborative filtering weave information tapestry
#@David Goldberg,David A. Nichols,Brian M. Oki,Douglas B. Terry
#t1992
#cCommun. ACM
#index801
#id801
#%
#! tapestry experimental mail developed xerox palo alto research center predicated belief information filtering effective humans involved filtering process tapestry designed support content based filtering collaborative filtering entails people collaborating help perform filtering recording reactions documents read reactions called annotations accessed peoples filters tapestry intended handle incoming stream electronic documents serves mail filter repository components indexer document store annotation store filterer little box remailer appraiser reader browser tapestrys client server architecture various components tapestry query language described

#* grouplens applying collaborative filtering usenet news
#@Joseph A. Konstan,Bradley N. Miller,David Maltz,Jonathan L. Herlocker,Lee R. Gordon,John Riedl
#t1997
#cCommun. ACM
#index802
#id802
#%
#! grouplens project designed implemented evaluated collaborative filtering usenet newsa high volume high turnover discussion list service internet usenet newsgroupsthe individual discussion listsmay carry hundreds messages day theory newsgroup organization allows readers select content interests practice

#* collaborative filtering gaussian probabilistic latent semantic analysis
#@Thomas Hofmann
#t2003
#cSIGIR
#index803
#id803
#%801,804,802
#! collaborative filtering aims learning predictive models user preferences interests behavior community data database available user preferences paper new model based algorithm designed task based generalization probabilistic latent semantic analysis continuous valued response variables specifically assume observed user ratings modeled mixture user communities groups users participate probabilistically groups community characterized gaussian distribution normalized ratings item normalization ratings performed user specific manner account variations absolute shift variance ratings experiments eachmovie data set proposed approach compares favorably collaborative filtering techniques

#* grouplens open architecture collaborative filtering netnews
#@Paul Resnick,Neophytos Iacovou,Mitesh Suchak,Peter Bergstrom,John Riedl
#t1994
#cCSCW
#index804
#id804
#%
#! collaborative filters help people make choices based opinions people grouplens collaborative filtering netnews help people articles like huge stream available articles news reader clients display predicted scores make easy users rate articles read rating servers called better bit bureaus gather disseminate ratings rating servers predict scores based heuristic people

#* collaborative filtering decoupled models preferences ratings
#@Rong Jin,Luo Si,ChengXiang Zhai,James P. Callan
#t2003
#cCIKM
#index805
#id805
#%806,807,804,808,809,810,811,812,813
#! paper new model collaborative filtering motivation work comes fact users similar preferences items different rating schemes example user tend assign higher rating items user unlike previous models collaborative filtering determine similarity users based rating performance model treats users preferences items separately users rating scheme specifically user build separate models preference model capturing items favored user rating model capturing user rate item given preference information similarity users computed based underlying preference model instead surface ratings compare new model representative previous approaches data sets experiment results new model outperforms previous approaches tested consistently data sets

#* empirical analysis predictive algorithms collaborative filtering
#@
#t0
#c
#index806
#id806
#%
#! collaborative filtering recommender systems use database user preferences predict additional topics products new user like paper algorithms designed task including techniques based correlation coefficients vector based similarity calculations statistical bayesian methods compare predictive accuracy various methods set representative problem domains use basic classes evaluation metrics characterizes accuracy set individual predictions terms average absolute deviation second estimates utility ranked list suggested items metric uses estimate probability user recommendation ordered list experiments run datasets associated application areas experimental protocols evaluation metrics various algorithms results indicate wide range conditions bayesian networks decision trees node correlation methods outperform bayesian clustering vector similarity methods correlation bayesian networks preferred method depends nature dataset nature application ranked versus presentation availability votes make predictions considerations include size database speed predictions learning time

#* collaborative filtering generalized vector space model
#@Ian Soboroff,Charles K. Nicholas
#t2000
#cSIGIR
#index807
#id807
#%
#! collaborative filtering technique recommending documents users based similar tastes users users tend agree like recommend documents generalized vector space model information retrieval represents document vector similarities documents process collaborative filtering nearly identical process retrieval using gvsm matrix user ratings using observation model filtering collaboratively using document content possible

#* latent class models collaborative filtering
#@
#t0
#c
#index808
#id808
#%
#! paper presents statistical approach collaborative filtering investigates use latent class models predicting individual choices preferences based observed preference behavior models discussed compared aspect model probabilistic latent space model models individual preferences convex combination preference factors sided clustering model simultaneously partitions persons objects clusters present em algorithms different variants aspect model derive approximate em algorithm based variational principle sided clustering model benefits different models experimentally investigated large movie data set

#* combining content collaboration text filtering
#@Ian Soboroff, Charles K. Nicholas
#t1999
#cProceedings of the IJCAI99 Workshop on Machine Learning for Information Filtering
#index809
#id809
#%
#! technique combining collabfi orative input document content text ltering technique uses latent semanfi tic indexing create collaborative view collection user pro les pro les selves term vectors constructed docfi uments deemed relevant user informafi tion need using standard text collections approach performs quite favorably compared contentfibased approaches

#* content boosted collaborative filtering improved recommendations
#@Prem Melville,Raymond J. Mooney,Ramadass Nagarajan
#t2002
#cAAAI/IAAI
#index810
#id810
#%
#! recommender systems use collaborative filtering content based methods predict new items user methods advantages individually fail provide good recommendations situations incorporating components methods hybrid recommender overcome shortcomings paper present elegant effective framework combining content collaboration approach uses content based predictor tc enhance existing user data provides personalized suggestions collaborative filtering present experimental results approach content boosted collaborative filtering performs better pure content based predictor pure collaborative filter naive hybrid approach

#* swami framework collaborative filtering algorithm development evaluation
#@Danyel Fisher,Kris Hildrum,Jason I. Hong,Mark W. Newman,Megan Thomas,Rich Vuduc
#t2000
#cSIGIR
#index811
#id811
#%
#! present java based framework swami shared wisdom amalgamation interpretations building studying collaborative filtering systems swami consists components prediction engine evaluation visualization component prediction engine provides common interface implementing different prediction algorithms evaluation provides standardized testing methodology metrics analyzing accuracy run time performance prediction algorithms visualization component suggests graphical representations inform development analysis prediction algorithms demonstrate swami eachmovie data set comparing prediction algorithms traditional pearson correlation based method support vector machines new accurate scalable correlation based method based clustering techniques

#* maximum likelihood incomplete data em algorithm
#@A. P. Dempster,N. M. Laird,D. B. Rubin
#t1977
#cJournal of the Royal Statistical Society
#index812
#id812
#%
#! broadly applicable algorithm computing maximum likelihood estimates incomplete data presented various levels generality theory showing monotone behaviour likelihood convergence algorithm derived examples sketched including missing value situations applications grouped censored truncated data finite mixture models variance component estimation hyperparameter estimation iteratively reweighted squares factor analysis

#* collaborative filtering personality diagnosis hybrid memory model based approach
#@DM Pennock,Eric Horvitz,S Lawrence,C. Lee Giles
#t2000
#cUAI
#index813
#id813
#%
#! growth internet commerce stimulated use collaborative filtering cf algorithms recommender systems systems leverage knowledge known preferences multiple users recommend items users cf methods harnessed make recommendations items web pages movies books toys researchers proposed evaluated approaches generating recommendations evaluate new method called personality diagnosis pd given users preferences items compute probability personality type users turn probability like new items pd retains advantages traditional similarity weighting techniques data brought bear prediction new data added easily incrementally additionally pd meaningful probabilistic interpretation leveraged justify explain augment results report empirical results eachmovie database movie ratings user profile data collected citeseer digital library science research papers probabilistic framework naturally supports variety descriptive measurementsin particular consider applicability value information voi computation

#* effective missing data prediction collaborative filtering
#@Hao Ma,Irwin King,Michael R. Lyu
#t2007
#cSIGIR
#index814
#id814
#%815,806,816,817,818,819,803,820,821,822,823,824,813,825,804,826,827,800,828,829
#! memory based collaborative filtering algorithms widely adopted popular recommender systems approaches suffer data sparsity poor prediction quality problems usually user item matrix quite sparse directly leads inaccurate recommendations paper focuses memory based collaborative filtering problems crucial factors similarity computation users items missing data prediction algorithms use enhanced pearson correlation coefficient pcc algorithm adding parameter overcomes potential decrease accuracy computing similarity users items second propose effective missing data prediction algorithm information users items taken account algorithm set similarity threshold users items respectively prediction algorithm determine predicting missing data address predict missing data employing combination user item information finally empirical studies dataset movielens shown newly proposed method outperforms stateofthe art collaborative filtering algorithms robust data sparsity

#* unifying collaborative content based filtering
#@Justin Basilico,Thomas Hofmann
#t2004
#cICML
#index815
#id815
#%
#! collaborative content based filtering paradigms applied context recommender systems user preference prediction paper proposes novel unified approach systematically integrates available training information past user item ratings attributes items users learn prediction function key ingredient method design suitable kernel similarity function user item pairs allows simultaneous generalization user item dimensions propose line algorithm jrank generalizes perceptron learning experimental results eachmovie data set significant improvements standard approaches

#* collaborative filtering privacy factor analysis
#@John F. Canny
#t2002
#cSIGIR
#index816
#id816
#%
#! collaborative filtering cf valuable commerce direct recommendations music movies news today systems disadvantages including privacy risks ubiquitous computing great potential individuals share kinds information places things buy privacy risks severe paper new method collaborative filtering protects privacy individual data method based probabilistic factor analysis model privacy protection provided peer peer protocol described outlined paper factor analysis approach handles missing data requiring default values experiments suggest accurate method cf date new algorithm advantages speed storage previous algorithms finally suggest applications approach kinds statistical analyses survey questionaire data

#* combining content based collaborative filters online newspaper
#@Mark Claypool,Anuja Gokhale,Tim Miranda,Pavel Murnikov,Dmitry Netes,Matthew Sartin
#t1999
#cProceedings of ACM SIGIR Workshop on Recommender Systems
#index817
#id817
#%
#! explosive growth mailing lists web sites usenet news demands effective filtering solutions collaborative filtering combines informed opinions humans make personalized accurate predictions content based filtering uses speed computers make complete fast predictions work present new filtering approach combines coverage speed content filters depth collaborative filtering apply research approach online newspaper untapped opportunity filters useful wide spread news reading populace present design filtering results preliminary experiments suggest merits approach

#* item based recommendation
#@M Deshpande,G Karypis
#t2004
#cACM Transactions on Information Systems
#index818
#id818
#%
#! explosive growth world wide web emergence commerce led development recommender systems personalized information filtering technology used identify set items certain user user based collaborative filtering successful technology building recommender systems date extensively used commercial recommender systems unfortunately computational complexity methods grows linearly number customers typical commercial applications millions address scalability concerns model based recommendation techniques developed techniques analyze user item matrix discover relations different items use relations compute list recommendations article present class model based recommendation algorithms determines similarities various items uses identify set items recommended key steps class algorithms method used compute similarity items ii method used combine similarities order compute similarity basket items candidate recommender item experimental evaluation real datasets shows item based algorithms orders magnitude faster traditional user neighborhood based recommender systems provide recommendations comparable better quality

#* empirical analysis design choices neighborhood based collaborative filtering algorithms
#@Jonathan L. Herlocker,Joseph A. Konstan,John Riedl
#t2002
#cInf. Retr.
#index819
#id819
#%
#! collaborative filtering systems predict user new items based recommendations people similar interests instead performing content indexing content analysis collaborative filtering systems rely entirely ratings members participating community predictions based human ratings collaborative filtering systems potential provide filtering based complex attributes quality taste aesthetics implementations collaborative filtering apply variation neighborhood based prediction algorithm variations similarity metrics weighting approaches combination measures rating normalization appeared implementation parameters consensus choice technique appropriate situations significant effect accuracy parameter consequently person implementing collaborative filtering make hard design choices little guidance article provides set recommendations guide design neighborhood based prediction systems based results empirical study apply analysis framework divides neighborhood based prediction approach components examines variants key parameters component components identified similarity computation neighbor selection rating combination

#* latent semantic models collaborative filtering
#@Thomas Hofmann
#t2004
#cACM Trans. Inf. Syst.
#index820
#id820
#%
#! collaborative filtering aims learning predictive models user preferences interests behavior community data database available user preferences article new family model based algorithms designed task algorithms rely statistical modelling technique introduces latent class variables mixture model setting discover user communities prototypical profiles investigate variations deal discrete continuous response variables different objective functions main advantages technique standard memory based methods higher accuracy constant time prediction explicit compact model representation used user communitites experimental evaluation shows substantial improvements accucracy existing methods published results obtained

#* automatic weighting scheme collaborative filtering
#@Rong Jin,Joyce Y. Chai,Luo Si
#t2004
#cSIGIR
#index821
#id821
#%
#! collaborative filtering identifies information particular user based information provided similar users memory based approaches collaborative filtering pearson correlation coefficient approach identify similarity users comparing ratings set items approaches different items weighted equally predefined functions impact rating discrepancies different users taken consideration example item highly favored users smaller impact user similarity item different types users tend different ratings simple weighting methods variance weighting try address problem empirical studies shown ineffective improving performance collaborative filtering paper present optimization algorithm automatically compute weights different items based ratings training users specifically new weighting scheme create clustered distribution user vectors item space bringing users similar interests closer separating users different interests distant empirical studies datasets shown new weighting scheme substantially improves performance pearson correlation coefficient method collaborative filtering

#* clustering collaborative filtering applications
#@Arnd Kohrs,Bernard Merialdo
#t1995
#cComputational Intelligence for Modelling, Control & Automation. IOS
#index822
#id822
#%
#! collaborative filtering systems assist users identify items providing predictions based ratings users quality predictions depends strongly available ratings collaborative filtering algorithms perform poorly ratings available paper identify important situations sparse ratings bootstrapping collaborative filtering users providing recommendations new users rated

#* amazon com recommendations item item collaborative filtering
#@Greg Linden,Brent Smith,Jeremy York
#t2003
#cIEEE Internet Computing
#index823
#id823
#%
#! recommendation algorithms best known use commerce web sites use input customers interests generate list recommended items applications use items customers purchase explicitly rate represent interests use attributes including items viewed demographic data subject interests favorite artists amazon com use recommendation algorithms personalize online store customer store radically changes based customer interests showing programming titles software engineer baby toys new mother click conversion rates important measures web based email advertising effectiveness vastly exceed untargeted content banner advertisements seller lists

#* collaborative filtering algorithm evaluation metric accurately model user experience
#@Matthew R. McLaughlin,Jonathan L. Herlocker
#t2004
#cSIGIR
#index824
#id824
#%
#! collaborative filtering cf systems researched decade tool deal information overload heart systems algorithms generate predictions recommendations article empirically demonstrate acclaimed cf recommendation algorithms flaws result dramatically unacceptable user experience response introduce new belief distribution algorithm overcomes flaws provides substantially richer user modeling belief distribution algorithm retains qualities nearest neighbor algorithms performed past produces predictions belief distributions rating values point rating value addition illustrate exclusive use mean absolute error metric concealed flaws long propose use modified precision metric accurately evaluating user experience

#* fast maximum margin matrix factorization collaborative prediction
#@Jason D. M. Rennie,Nathan Srebro
#t2005
#cICML
#index825
#id825
#%
#! maximum margin matrix factorization mmmf recently suggested srebro et al convex infinite dimensional alternative low rank approximations standard factor models mmmf formulated semi definite programming sdp learned using standard sdp solvers current sdp solvers handle mmmf problems matrices dimensionality investigate direct gradient based optimization method mmmf demonstrate large collaborative prediction problems compare results obtained marlin mmmf substantially outperforms methods tested

#* item based collaborative filtering recommendation algorithms
#@Badrul M. Sarwar,G Karypis,Joseph A. Konstan,John Riedl
#t2001
#cWWW
#index826
#id826
#%830,831,832,806,801,833,834,802,804,835,836,837,838,800
#! recommender systems apply knowledge discovery techniques problem making personalized recommendations information products services live interaction systems especially nearest neighbor collaborative filtering based ones achieving widespread success web tremendous growth available information number visitors web sites recent years poses key challenges recommender systems producing high quality recommendations performing recommendations second millions users items achieving high coverage face data sparsity traditional collaborative filtering systems work increases number participants new recommender technologies needed quickly produce high quality recommendations large scale problems address issues explored item based collaborative filtering techniques item based techniques analyze user item matrix identify relationships different items use relationships indirectly compute recommendations users paper analyze different item based recommendation generation algorithms look different techniques computing item item similarities item item correlation vs cosine similarities item vectors different techniques obtaining recommendations weighted sum vs regression model finally experimentally evaluate results compare basic nearest neighbor approach experiments suggest item based algorithms provide dramatically better performance user based algorithms time providing better quality best available userbased algorithms

#* flexible mixture model collaborative filtering
#@Luo Si,Rong Jin
#t2003
#cICML
#index827
#id827
#%
#! paper presents flexible mixture model fmm collaborative filtering fmm extends existing partitioning clustering algorithms collaborative filtering clustering users items simultaneously assuming user item belong single cluster furthermore introduction preference nodes proposed framework able explicitly model users rate items vary dramatically users similar tastes items empirical study datasets movie ratings shown new algorithm outperforms collaborative filtering algorithms substantially

#* unifying user based item based collaborative filtering approaches similarity fusion
#@Jun Wang,Arjen P. de Vries,Marcel J. T. Reinders
#t2006
#cSIGIR
#index828
#id828
#%
#! memory based methods collaborative filtering predict new ratings averaging weighted ratings respectively pairs similar users items practice large number ratings similar users similar items available sparsity inherent rating data consequently prediction quality poor paper reformulates memory based collaborative filtering problem generative probabilistic framework treating individual user item ratings predictors missing ratings final rating estimated fusing predictions sources predictions based ratings item users predictions based different item ratings user ratings predicted based data similar users rating similar items existing user based item based approaches correspond simple cases framework complete model robust data sparsity different types ratings used concert additional ratings similar users similar items employed background model smooth predictions experiments demonstrate proposed methods robust data sparsity better recommendations

#* scalable collaborative filtering using cluster based smoothing
#@Gui-Rong Xue,Chenxi Lin,Qiang Yang,Wensi Xi,Hua-Jun Zeng,Yong Yu,Zheng Chen
#t2005
#cSIGIR
#index829
#id829
#%839,840,806,817,811,841,842,843,808,822,813,804,807,800,844
#! memory based approaches collaborative filtering identify similarity users comparing ratings set items past memory based approaches shown suffer fundamental problems data sparsity difficulty scalability alternatively model based approaches proposed alleviate problems approaches tends limit range users paper present novel approach combines advantages kinds approaches introducing smoothing based method approach clusters generated training data provide basis data smoothing neighborhood selection result provide higher accuracy increased efficiency recommendations empirical studies datasets eachmovie movielens new proposed approach consistently outperforms state art collaborative filtering algorithms

#* horting hatches egg new graph theoretic approach collaborative filtering
#@Charu C. Aggarwal,Joel L. Wolf,Kun-Lung Wu,Philip S. Yu
#t1999
#cKDD
#index830
#id830
#%
#! paper introduces new novel approach ratingbased collaborative filtering new technique appropriate commerce merchants offering groups relatively homogeneous items compact disks videos books software like contrast known collaborative filtering techniques new algorithm graph theoretic based twin new concepts ho rag predictability demonstrated paper technique fast scalable accurate requires modest learning curve makes use hierarchical classification scheme order introduce context rating process uses called creative links order surprising atypical items recommend items cross group boundaries new technique key engines intelligent recommendation algorithm ira project developed ibm research addition recommendation engines ira contains situation analyzer determine appropriate mix engines particular commerce merchant engine optimizing placement advertisements

#* recommendation classification using social content based information recommendation
#@Chumki Basu,Haym Hirsh,William W. Cohen
#t1998
#cAAAI/IAAI
#index831
#id831
#%
#! recommendation systems make suggestions artifacts user instance predict user interested seeing particular movie social recomendation methods collect ratings artifacts individuals use nearest neighbor techniques make recommendations user concerning new artifacts methods use significant information available nature artifact cast lists

#* learning collaborative information filters
#@D Billsus,M Pazzani
#t1998
#cICML
#index832
#id832
#%
#! predicting items user like basis users ratings items established strategy adopted recommendation services internet seen classification problem algorithms proposed far draw results machine learning literature propose representation collaborative filtering tasks allows application virtually machine learning algorithm identify shortcomings current collaborative filtering techniques propose use learning algorithms paired feature extraction techniques specifically address limitations previous approaches best performing algorithm based singular value decomposition initial matrix user ratings exploiting latent structure essentially eliminates need users rate common items order predictors preferences evaluate proposed algorithm large database user ratings motion pictures approach significantly outperforms current collaborative filtering algorithms

#* combining collaborative filtering personal agents better recommendations
#@Nathaniel Good,J. Ben Schafer,Joseph A. Konstan,Al Borchers,Badrul M. Sarwar,Jonathan L. Herlocker,John Riedl
#t1999
#cAAAI/IAAI
#index833
#id833
#%
#! information filtering agents collaborative filtering attempt alleviate information overload identifying items user worthwhile information filtering focuses analysis item content development personal user profile collaborative filtering cf focuses identification users similar tastes use opinions recommend items technique advantages limitations suggest beneficially combined paper shows cf framework used combine personal agents opinions community users produce better recommendations agents users produce shows using cf create personal combination set agents produces better results individual agents combination mechanisms key implication results users avoid having select agents use let cf framework select best ones

#* evaluation item based recommendation algorithms
#@G Karypis
#t2001
#cCIKM
#index834
#id834
#%
#! explosive growth world wide web emergence commerce led development recommender systems personalized information filtering technology used identify set items certain user user based collaborative filtering successful technology building recommender systems date extensively used commercial recommender systems unfortunately computational complexity methods grows

#* recommender systems
#@Hannes Werthner,Hans Robert Hansen,Francesco Ricci
#t2007
#cHICSS
#index835
#id835
#%
#! recommender systems assist augment natural social process typical recommender people provide recommendations inputs aggregates directs appropriate recipients cases primary

#* using filtering agents improve prediction quality grouplens research collaborative filtering
#@Badrul M. Sarwar,Joseph A. Konstan,Al Borchers,Jonathan L. Herlocker,Bradley N. Miller,John Riedl
#t1998
#cCSCW
#index836
#id836
#%
#! collaborative filtering systems help address information overload using opinions users community make personal recommendations documents user collaborative filtering systems user opinions relative large number documents available sparsity problem reduce utility filtering reducing number documents make recommendations adversely affecting quality recommendations paper defines implements model integrating content based ratings collaborative filtering filterbot model allows collaborative filtering systems address sparsity tapping strength content filtering techniques identify evaluate metrics assessing effectiveness filterbots specifically filtering enhancements general finally experimentally validate filterbot approach showing simple filterbots spell checking increase utility users sparsely populated collaborative filtering systems

#* recommender systems commerce
#@J. Ben Schafer,Joseph A. Konstan,John Riedl
#t1999
#cACM Conference on Electronic Commerce
#index837
#id837
#%
#! recommender systems changing novelties used commerce sites business tools shaping world commerce largest commerce web sites using recommender systems help customers products purchase recommender learns customer recommends products valuable available products paper present explanation recommender systems help commerce sites increase sales analyze sites use recommender systems including sites use recommender based examples create taxonomy recommender systems including interfaces present customers technologies used create recommendations inputs need customers conclude ideas new applications recommender systems commerce

#* phoaks sharing recommendations
#@
#t0
#cnull
#index838
#id838
#%
#! finding relevant high quality information theworld wide web difficult problem phoaks people helping know stuff experimental addresses problem collaborative filtering approach phoaks works automatically recognizing tallying redistributing recommendations web resources mined usenet news messages

#* fab content based collaborative recommendation
#@Marko Balabanovic,Yoav Shoham
#t1997
#cCommun. ACM
#index839
#id839
#%
#! online readers need tools help cope mass content available world wide web traditional media readers provided assistance making selections includes implicit assistance form editorial oversight explicit assistance form recommendation services movie reviews restaurant guides electronic medium offers new opportunities create recommendation services ones adapt time track evolving interests fab recommendation web operational versions december

#* class based gram models natural language
#@Peter F. Brown,Vincent J. Della Pietra,Peter V. de Souza,Jennifer C. Lai,Robert L. Mercer
#t1992
#cComputational Linguistics
#index840
#id840
#%
#! address problem predicting word previous words sample text particular discuss gram models based classes words discuss statistical algorithms assigning words classes based frequency occurrence words able extract classes flavor syntactically based groupings semantically based groupings depending nature underlying statistics

#* eigentaste constant time collaborative filtering algorithm
#@Kenneth Y. Goldberg,Theresa Roeder,Dhruv Gupta,Chris Perkins
#t2001
#cInf. Retr.
#index841
#id841
#%
#! eigentaste collaborative filtering algorithm uses universal queries elicit real valued user ratings common set items applies principal component analysis pca resulting dense subset ratings matrix pca facilitates dimensionality reduction offline clustering users rapid computation recommendations database users standard nearest neighbor techniques require processing time compute recommendations eigentaste requires constant time compare eigentaste alternative algorithms using data jester online joke recommending jester collected approximately ratings users use normalized mean absolute error nmae measure compare performance different algorithms appendix use uniform normal distribution models derive analytic estimates nmae predictions random jester dataset eigentaste computes recommendations orders magnitude faster loss accuracy jester online http eigentaste berkeley edu

#* evaluating collaborative filtering recommender systems
#@Jonathan L. Herlocker,Joseph A. Konstan,Loren G. Terveen,John Riedl
#t2004
#cACM Trans. Inf. Syst.
#index842
#id842
#%
#! recommender systems evaluated incomparable ways article review key decisions evaluating collaborative filtering recommender systems user tasks evaluated types analysis datasets used ways prediction quality measured evaluation prediction attributes quality user based evaluation addition reviewing evaluation strategies used prior researchers present empirical results analysis various accuracy metrics content domain tested metrics collapsed roughly equivalence classes metrics equivalency class strongly correlated metrics different equivalency classes uncorrelated

#* applying associative retrieval techniques alleviate sparsity problem collaborative filtering
#@Zan Huang,Hsinchun Chen,Daniel Dajun Zeng
#t2004
#cACM Trans. Inf. Syst.
#index843
#id843
#%
#! recommender systems widely applied application settings suggest products services information items potential consumers collaborative filtering successful recommendation approach makes recommendations based past transactions feedback consumers sharing similar interests major problem limiting usefulness collaborative filtering sparsity problem refers situation transactional feedback data sparse insufficient identify similarities consumer interests article propose deal sparsity problem applying associative retrieval framework related spreading activation algorithms explore transitive associations consumers past transactions feedback transitive associations valuable source information help infer consumer interests explored deal sparsity problem evaluate effectiveness approach conducted experimental study using data set online bookstore experimented spreading activation algorithms including constrained leaky capacitor algorithm branch bound serial symbolic search algorithm hopfield net parallel relaxation search algorithm algorithms compared collaborative filtering approaches consider transitive associations simple graph search approach variations user based approach item based approach experimental results indicate spreading activation based approaches significantly outperformed collaborative filtering methods measured recommendation precision recall measure rank score observed activation effect spreading activation approach incorporating transitive associations past transactional data sparse dilute data used infer user preferences lead degradation recommendation performance

#* similarity measure instance selection collaborative filtering
#@Chun Zeng,Chun-Xiao Xing,Li-Zhu Zhou
#t2003
#cWWW
#index844
#id844
#%901,865,902,903,802,864,839,817,904,806,905,804,906,907,808
#! collaborative filtering successful research applications information filtering commerce nearest neighbor knn method popular way realization key technique nearest neighbors given user predict interests method suffers fundamental problems sparsity scalability paper present solutions problems adopt techniques matrix conversion method similarity measure instance selection method present improved collaborative filtering algorithm based methods contrast existing collaborative algorithms method shows satisfactory accuracy performance

#* eigenrank ranking oriented approach collaborative filtering
#@Nathan N. Liu
#t2008
#cSIGIR '08
#index845
#id845
#%806,846,847,848,849,841,850,819,820,851,852,805,853,802,823,814,854,804,855,826,829
#! recommender able suggest items likely preferred user systems degree preference represented rating score given database users past ratings set items traditional collaborative filtering algorithms based predicting potential ratings user assign unrated items ranked predicted ratings produce list recommended items paper propose collaborative filtering approach addresses item ranking problem directly modeling user preferences derived ratings measure similarity users based correlation rankings items rating values propose new collaborative filtering algorithms ranking items based preferences similar users experimental results real world movie rating data sets proposed approach outperforms traditional collaborative filtering algorithms significantly ndcg measure evaluating ranked results

#* learning rank using gradient descent
#@Christopher J. C. Burges,Tal Shaked,Erin Renshaw,Ari Lazier,Matt Deeds,Nicole Hamilton,Gregory N. Hullender
#t2005
#cICML
#index846
#id846
#%
#! investigate using gradient descent methods learning ranking functions propose simple probabilistic cost function introduce ranknet implementation ideas using neural network model underlying ranking function present test results toy data data commercial internet search engine

#* learning order things
#@William W. Cohen,Robert E. Schapire,Yoram Singer
#t1999
#cJ. Artif. Intell. Res. (JAIR)
#index847
#id847
#%
#! applications desirable order classify instances consider problem learning order given feedback form preference judgments statements effect instance ranked ahead outline stage approach learns conventional means preference function form preffi indicates advisable rank new instances ordered maximize agreements learned preference function problem finding ordering agrees best preference function np complete restrictive assumptions simple greedy algorithm guaranteed good approximation discuss line learning algorithm based hedge algorithm finding good linear combination ranking experts use ordering algorithm combined line learning algorithm combination search experts domain specific query expansion strategy www search engine present experimental results demonstrate merits approach

#* efficient boosting algorithm combining preferences
#@Yoav Freund,Raj D. Iyer,Robert E. Schapire,Yoram Singer
#t2003
#cJournal of Machine Learning Research
#index848
#id848
#%
#! study problem learning accurately rank set objects combining given collection ranking preference functions problem combining preferences arises applications combining results different search engines collaborativefiltering problem ranking movies user based movie rankings provided users work begin presenting formal framework general problem analyze efficient algorithm called rankboost combining preferences based boosting approach machine learning theoretical results describing algorithms behavior training data new test data seen training efficient implementation algorithm particular restricted common case discuss experiments carried assess performance rankboost experiment used algorithm combine different web search strategies query expansion given domain second experiment collaborative filtering task making movie recommendations

#* computers intractability guide theory np completeness
#@
#t0
#cnull
#index849
#id849
#%
#! book opens description np completeness useful concept turns difficult prove given problem guaranteed long time solve class problems ones np complete provably equivalently difficult solve good algorithm defined runs polynomial time probably worth working finding good algorithm solve problem exactly smart people tried failed

#* topic sensitive pagerank
#@T. Haveliwala 
#t0
#cnull
#index850
#id850
#%856,857,858,859,860,861,852,175,669,855
#! original pagerank algorithm improving ranking search query results single pagerank vector computed using link structure web capture relative importance ofweb pages independent particular search query yield accurate search results propose computing set pagerank vectors biased using set representative topics capture accurately notion importance respect particular topic using precomputed biased pagerank vectors generate query specific importance scores pages query time generate accurate rankings single generic pagerank vector ordinary keyword search queries compute topic sensitive pagerank scores pages satisfying query using topic query keywords searches context search query performed highlighting words web page compute topic sensitive pagerank scores using topic context query appeared

#* cumulated gain based evaluation ir techniques
#@Kalervo Jarvelin,Jaana Kekalainen
#t2002
#cACM Trans. Inf. Syst.
#index851
#id851
#%
#! modern large retrieval environments tend overwhelm users large output documents equal relevance users highly relevant documents identified ranked presentation order develop ir techniques direction necessary develop evaluation approaches methods credit ir methods ability retrieve highly relevant documents extending traditional evaluation methods recall precision based binary relevance judgments graded relevance judgments alternatively novel measures based graded relevance judgments developed article proposes novel measures compute cumulative gain user obtains examining retrieval result given ranked position accumulates relevance scores retrieved documents ranked result list second similar applies discount factor relevance scores order devaluate late retrieved documents computes relative ideal performance ir techniques based cumulative gain able yield novel measures defined discussed use demonstrated case study using trec data sample run results queries trec relevance base used novel graded relevance judgments point scale test results indicate proposed measures credit ir methods ability retrieve highly relevant documents allow testing statistical significance effectiveness differences graphs based measures provide insight performance ir techniques allow interpretation example user point view

#* scaling personalized web search
#@Glen Jeh,Jennifer Widom
#t2003
#cWWW
#index852
#id852
#%
#! recent web search techniques augment traditional text matching global notion importance based linkage structure web google pagerank algorithm refined searches global notion importance specialized create personalized views importance example importance scores biased according user specified set initially interesting pages computing storing possible personalized views advance impractical computing personalized views query time computation view requires iterative computation web graph present new graph theoretical results new technique based results encode personalized views partial vectors partial vectors shared multiple personalized views computation storage costs scale number views approach enables incremental computation construction personalized views partial vectors practical query time present efficient dynamic programming algorithms computing partial vectors algorithm constructing personalized views partial vectors experimental results demonstrating effectiveness scalability techniques

#* optimizing search engines using clickthrough data
#@Thorsten Joachims
#t2002
#cKDD
#index853
#id853
#%848,864,865,866
#! paper presents approach automatically optimizing retrieval quality search engines using clickthrough data intuitively good information retrieval present relevant documents high ranking relevant documents following previous approaches learning retrieval functions examples exist typically require training data generated relevance judgments experts makes difficult expensive apply goal paper develop method utilizes clickthrough data training query log search engine connection log links users clicked presented ranking clickthrough data available abundance recorded low cost taking support vector machine svm approach paper presents method learning retrieval functions theoretical perspective method shown founded risk minimization framework furthermore shown feasible large sets queries features theoretical results verified controlled experiment shows method effectively adapt retrieval function meta search engine particular group users outperforming google terms retrieval quality couple training examples

#* collaborative filtering personality diagnosis hybrid memory model based approach
#@DM Pennock,Eric Horvitz,S Lawrence,C. Lee Giles
#t2000
#cUAI
#index854
#id854
#%
#! growth internet commerce stimulated use collaborative filtering cf algorithms recommender systems systems leverage knowledge known preferences multiple users recommend items users cf methods harnessed make recommendations items web pages movies books toys researchers proposed evaluated approaches generating recommendations evaluate new method called personality diagnosis pd given users preferences items compute probability personality type users turn probability like new items pd retains advantages traditional similarity weighting techniques data brought bear prediction new data added easily incrementally additionally pd meaningful probabilistic interpretation leveraged justify explain augment results report empirical results eachmovie database movie ratings user profile data collected citeseer digital library science research papers probabilistic framework naturally supports variety descriptive measurementsin particular consider applicability value information voi computation

#* intelligent surfer probabilistic combination link content information pagerank
#@
#t0
#cnull
#index855
#id855
#%
#! pagerank algorithm used google search engine greatly improves results web search taking account link structure web pagerank assigns page score proportional number times random surfer visit page surfed indefinitely page page following outlinks page equal probability propose improve pagerank using intelligent surfer guided probabilistic model relevance page query efficient execution algorithm query time possible precomputing crawl time queries necessary terms experiments large subsets web indicate algorithm significantly outperforms pagerank humanrated quality pages returned remaining efficient used todays large search engines

#* improved algorithms topic distillation hyperlinked environment
#@Krishna Bharat,Monika Rauch Henzinger
#t1998
#cSIGIR
#index856
#id856
#%867,858,868,868,869,175,870,871,669,872
#! paper addresses problem topic distillation theworld wideweb given typical user query quality documents related query topic connectivity analysis shown useful identifying high quality pages topic specific graph hyperlinked documents essence approach augment previous connectivity analysis based algorithm content analysis identify problems existing approach devise algorithms tackle results user evaluation reported improvement precision documents pure connectivity analysis

#* anatomy large scale hypertextual web search engine
#@
#t0
#cnull
#index857
#id857
#%
#! paper present google prototype large scale search engine makes heavy use structure present hypertext google designed crawl index web efficiently produce satisfying search results existing systems prototype text hyperlink database million pages available http google stanford edu engineer search engine challenging task search engines index tens hundreds millions web pages involving comparable number distinct terms answer tens millions queries day despite importance large scale search engines web little academic research furthermore rapid advance technology web proliferation creating web search engine today different years ago paper provides depth description large scale web search engine detailed public description know date apart problems scaling traditional search techniques data magnitude new technical challenges involved using additional information present hypertext produce better search results paper addresses question build practical large scale exploit additional information present hypertext look problem effectively deal uncontrolled hypertext collections publish want

#* automatic resource compilation analyzing hyperlink structure associated text
#@Soumen Chakrabarti,Byron Dom,Prabhakar Raghavan,S. Rajagopalan,D. Gibson,Jon M. Kleinberg
#t1998
#cComputer Networks
#index858
#id858
#%
#! design prototyping evaluation arc automatically compiling list authoritative web resources sufficiently broad topic goal arc compile resource lists similar provided yahoo infoseek fundamental difference services construct lists manually combination human automated effort arc operates fully automatically evaluation arc yahoo infoseek resource lists panel human users evaluation suggests resources arc frequently fare better lists resources manually compiled classified topic provide examples arc resource lists reader examine

#* rank aggregation methods web
#@
#t0
#cnull
#index859
#id859
#%
#! consider problem combining ranking results various sources context web main applications include building meta search engines combining ranking functions selecting documents based multiple criteria improving search precision word associations develop set techniques rank aggregation problem compare performance known methods primary goal work design rank aggregation techniques ectively combat spam problem web searches experiments methods simple efficient ective

#* placing search context concept revisited
#@Lev Finkelstein,Evgeniy Gabrilovich,Yossi Matias,Ehud Rivlin,Zach Solan,Gadi Wolfman,Eytan Ruppin
#t2002
#cACM Trans. Inf. Syst.
#index860
#id860
#%
#! keyword based search engines widespread use today popular means web based information retrieval systems deceptively simple considerable skill required order satisfy non trivial information needs paper presents new conceptual paradigm performing search context largely automates search process providing non professional users highly relevant results paradigm implemented practice intellizap search initiated text query marked user document views guided text surrounding marked query document context context driven information retrieval process involves semantic keyword extraction clustering automatically generate new augmented queries submitted host general domain specific search engines search results semantically reranked using context experimental results testify using context guide search effectively offers inexperienced users advanced search tool web

#* efficient computation pagerank
#@T. Haveliwala
#t1999
#cTechnical Report
#index861
#id861
#%
#! paper discusses efficient techniques computing pagerank ranking metric hypertext documents pagerank computed large subgraphs web hundreds millions nodes machines limited main memory running time measurements various memory configurations presented pagerank computation million page stanford webbase archive discuss methods analyzing convergence pagerank based induced ordering pages present convergence results helpful determining number iterations necessary achieve useful pagerank assignment absence presence search queries

#* webwatcher tour guide world wide web
#@Thorsten Joachims,Dayne Freitag,Tom Mitchell
#t1997
#cInternational Joint Conference on Artificial Intelligence
#index864
#id864
#%
#! explore notion tour guide software agent assisting users browsing world wide web web tour guide agent provides assistance similar provided human tour guide museum guides user appropriate path collection based knowledge user interests location relevance various items collection way interacted collection past paper describes simple operational tour guide called webwatcher given tours people browsing cmu school science web pages webwatcher accompanies users page page suggests appropriate hyperlinks learns experience improve advice giving skills learning algorithms used webwatcher experimental results showing effectiveness lessons learned case study web tour guide agents

#* letizia agent assists web browsing
#@Henry Lieberman
#t1995
#cIJCAI (1)
#index865
#id865
#%
#! letizia user interface agent assists user browsing world wide web user operates conventional web browser netscape agent tracks user behavior attempts anticipate items doing concurrent autonomous exploration links user current position agent automates browsing strategy consisting best search augmented heuristics inferring user browsing behavior introduction letizia lvarez

#* term weighting approaches automatic text retrieval
#@Gerard Salton,Chris Buckley
#t1988
#cInformation Processing and Management: an International Journal
#index866
#id866
#%
#! experimental evidence accumulated past years indicates textindexing systems based assignment appropriately weighted single terms produce retrieval results superior obtainable elaborate text representations results depend crucially choice effective term weighting systems paper summarizes insights gained automatic term weighting provides baseline single term indexing models elaborate content analysis procedures compared

#* connectivity server fast access linkage information web
#@Krishna Bharat,AZ Broder,Monika Rauch Henzinger,Puneet Kumar,Suresh Venkatasubramanian
#t1998
#cComputer Networks
#index867
#id867
#%
#! built server provides linkage information pages indexed altavista search engine basic operation server accepts query consisting set urls returns list pages point pages predecessors list pages pointed pages successors generally server produce entire neighbourhood graph theory sense given distance include information links exist pages neighbourhood information retrieved directly alta vista search engines engines optimized purpose process constructing neighbourhood given set pages slow laborious contrast prototype server needs ms result url far built applications use connectivity server direct interface permits fast navigation web predecessor successor relation visualization tool neighbourhood given set pages envisage numerous applications ranking visualization classification

#* search engines world wide web comparative study evaluation methodology
#@H Chu,M Rosenthal
#t1996
#cASIS 1996 Annual Conference Proceedings
#index868
#id868
#%
#! web search engines alta vista excite lycos compared evaluated terms search capabilities boolean logic truncation field search word phrase search retrieval performances precision response time using sample queries drawn real reference questions recall evaluation criterion information retrieval deliberately omitted study impossible assume relevant items particular query huge changing web authors study alta vista outperformed excite lycos search facilities retrieval performance lycos largest coverage web resources web search engines examined result research proposed methodology evaluating web search engines included current study

#* interactive query expansion
#@Donna Harman
#t1988
#cSIGIR
#index869
#id869
#%
#! era online retrieval appropriate offer guidance users wishing improve initial queries form guidance short lists suggested terms gathered feedback nearest neighbors term variants original query terms verify approach series experiments run using cranfield test collection discover techniques select terms lists effective retrieval results significant improvement expected approach query expansion

#* bibliometrics world wide web exploratory analysis intellectual structure cyberspace
#@Ray R. Larson
#t1996
#cProceedings American Society for Information Science
#index870
#id870
#%
#! exploratory study examines explosive growth bibliometrics world wide web based analysis gigabytes web pages collected inktomi web crawler use dec altavista search engine cocitation analysis set earth science related www sites statistical characteristics web documents hypertext links examined examination characteristics highly cited web documents

#* interfaces end user information seeking
#@Gary Marchionini
#t1992
#cJASIS
#index871
#id871
#%
#! discusses essential features interfaces support end user information seeking highlights include cognitive engineering task models task analysis problem solving nature information seeking examples systems end users including online public access catalogs opacs hypertext help systems suggested research directions

#* silk sow ear extracting usable structures web
#@Peter Pirolli,James E. Pitkow,Ramana Rao
#t1996
#cCHI
#index872
#id872
#%
#! current implementation world wide web lacks explicit structure strong typing closed hypertext systems property directly fueled explosive acceptance web complicates difficult problem identifying usable structures aggregates large hypertext collections reduced structures localities form basis simplifying visualizations navigation complex hypertext systems

#* learning block importance models web pages
#@Ruihua Song,Haifeng Liu,Ji-Rong Wen,Wei-Ying Ma
#t2004
#cWWW
#index873
#id873
#%874,857,875,876,877,878,879,880,881,882
#! previous works web page partitioned multiple segments blocks usually importance blocks page equivalent proved differentiating noisy unimportant blocks pages facilitate web mining search accessibility works uniform approach model presented measure importance different portions web pages user study people consistent view importance blocks web pages paper investigate model automatically assign importance values blocks web page define block importance estimation learning problem use vips vision based page segmentation algorithm partition web page semantic blocks hierarchy structure spatial features position size content features number images links extracted construct feature vector block based features learning algorithms svm neural network applied train various block importance models experiments best model achieve performance micro micro accuracy quite close persons

#* template detection data mining applications
#@Ziv Bar-Yossef,S. Rajagopalan
#t2002
#cWWW
#index874
#id874
#%
#! formulate propose template detection problem suggest practical solution based counting frequent item sets use templates pervasive web principles characterize assumptions hypertext information retrieval ir data mining dm systems templates major source violation principles consequence basic pure implementations simple search algorithms coupled template detection elimination surprising increases precision levels recall

#* function based object model website adaptation
#@Jinlin Chen,Baoyao Zhou,Jin Shi,HongJiang Zhang,Qiu Fengwu
#t2001
#cWWW
#index875
#id875
#%
#! content understanding crucial issue website adaptation paper present function based object model fom attempts understand authors intention identifying object function instead semantic understanding object website serves certain functions basic specific function reflect authors intention purpose object based consideration proposed fom model website understanding fom includes complementary parts basic fom based basic functional properties object specific fom based category object automatic approach detect functional properties category object presented fom generation level adaptation rules general rules specific rules based fom combined practical adaptation web content adaptation wireless application protocol wap developed application example proposed model experiments shown satisfactory results extensibility

#* geometrical statistical properties systems linear inequalities applications pattern recognition
#@TM Cover
#t1965
#cIEEE Transactions on Electronic Computers
#index876
#id876
#%
#! paper develops separating capacities families nonlinear decision surfaces direct application theorem classical combinatorial geometry shown family surfaces having degrees freedom natural separating capacity pattern vectors extending unifying results winder pattern separating capacity hyperplanes applying ideas vertices binary cube yields bounds number spherically quadratically general nonlinearly separable boolean functions variables

#* error correcting output codes general method improving multiclass inductive learning programs
#@Thomas G. Dietterich,Ghulum Bakiri
#t1991
#cAAAI
#index877
#id877
#%
#! multiclass learning problems involve finding definition unknown function range discrete set containing values classes definition acquired studying large collections training examples form xi xi existing approaches problem include direct application multiclass algorithms decision tree algorithms id cart application binary concept learning algorithms learn individual binary functions ic classes application binary concept learning algorithms distributed output codes employed sejnowski rosenberg nettalk paper compares approaches new technique bch error correcting codes employed distributed output representation output representations improve performance id nettalk task backpropagation isolated letter speech recognition task results demonstrate error correcting output codes provide general purpose method improving performance inductive learning programs multiclass problems

#* discovering informative content blocks web documents
#@Shian-Hua Lin,Jan-Ming Ho
#t2002
#cKDD
#index878
#id878
#%
#! paper propose new approach discover informative contents set tabular documents web pages web site infodiscoverer partitions page content blocks according html tag web page based occurrence features terms set pages calculates entropy value feature according entropy value feature content block entropy value block defined analyzing

#* automatic browsing large pictures mobile devices
#@Hao Liu,Xing Xie,Wei-Ying Ma,HongJiang Zhang
#t2003
#cACM Multimedia
#index879
#id879
#%
#! ictures increasingly common popular mobile communications limitation mobile devices need develop new technologies facilitate browsing large pictures small screen paper propose novel approach able automate scrolling navigation large picture minimal user interaction mobile devices image attention model employed illustrate information structure image optimal image browsing path calculated based image attention model simulate human browsing behaviors experimental evaluations proposed mechanism indicate approach effective way viewing large images small displays

#* web page cleaning web mining feature weighting
#@Lan Yi,Bing Liu
#t2003
#cIJCAI
#index880
#id880
#%
#! unlike conventional data text web pages typically contain large information main contents pages banner ads navigation bars copyright notices irrelevant information web page noise web pages seriously harm web mining clustering classification paper propose novel feature weighting technique deal web page noise enhance web mining method builds compressed structure tree capture common structure comparable blocks set web pages uses information based measure evaluate importance node compressed structure tree based tree node importance values method assigns weight word feature content block resulting weights used web mining evaluated proposed technique web mining tasks web page clustering web page classification experimental results weighting method able dramatically improve mining results

#* eliminating noisy information web pages data mining
#@Lan Yi,Bing Liu,Xiaoli Li
#t2003
#cKDD
#index881
#id881
#%
#! commercial web page typically contains information blocks apart main content blocks usually blocks navigation panels copyright privacy notices advertisements business purposes easy user access blocks main content blocks page noisy blocks information contained noisy blocks seriously harm web data mining eliminating noises great importance paper propose noise elimination technique based following observation given web site noisy blocks usually share common contents presentation styles main content blocks pages diverse actual contents presentation styles based observation propose tree structure called style tree capture common presentation styles actual contents pages given web site sampling pages site style tree built site site style tree sst introduce information based measure determine parts sst represent noises parts represent main contents site sst employed detect eliminate noises web page site mapping page sst proposed technique evaluated data mining tasks web page clustering classification experimental results noise elimination technique able improve mining results significantly

#* improving pseudo relevance feedback web information retrieval using web page segmentation
#@Shipeng Yu,Deng Cai,Ji-Rong Wen,Wei-Ying Ma
#t2003
#cWWW
#index882
#id882
#%
#! contrast traditional document retrieval web page good information unit search contains multiple topics lot irrelevant information navigation decoration interaction page paper propose vision based page segmentation vips algorithm detect semantic content structure web page compared simple dom based segmentation method page segmentation scheme utilizes useful visual cues obtain better partition page semantic level using vips algorithm assist selection query expansion terms pseudo relevance feedback web information retrieval achieve performance improvement web track dataset

#* link fusion unified link analysis framework multi type interrelated data objects
#@Wensi Xi,Benyu Zhang,Zheng Chen,Yizhou Lu,Shuicheng Yan,Wei-Ying Ma,Edward A. Fox
#t2004
#cWWW
#index883
#id883
#%856,857,858,884,885,886,887,888,889,890,891,175
#! web link analysis proven significant enhancement quality based web search existing links classified categories intra type links web hyperlinks represent relationship data objects homogeneous data type web pages inter type links user browsing log represent relationship data objects different data types users web pages unfortunately link analysis research considers type link paper propose unified link analysis framework called link fusion considers inter intra type link structure multiple type inter related data objects brings order objects data type time pagerank hits algorithms shown special cases unified link analysis framework experiments instantiation framework makes use user data web pages extracted proxy log proposed algorithm improve search effectiveness hits directhit algorithms respectively

#* mining web link structure
#@Soumen Chakrabarti,Byron Dom,Ravi Kumar,Prabhakar Raghavan,S. Rajagopalan,Andrew Tomkins,D. Gibson,Jon M. Kleinberg
#t1999
#cIEEE Computer
#index884
#id884
#%
#! web hypertext body approximately million pages continues grow roughly million pages day page variation prodigious data raw scale taken set web pages lacks unifying structure shows far authoring style content variation seen traditional text document collections level complexity makes shelf database management information retrieval solution impossible date index based search engines web primary tool users search information engines build giant indices let quickly retrieve set web pages containing given word string experienced users make effective use engines tasks solved searching tightly constrained keywords phrases search engines unsuited wide range equally important tasks particular topic breadth typically contain thousand million relevant web pages sea pages search engine select correct ones value user

#* learning probabilistically identify authoritative documents
#@David Cohn,Huan Chang
#t2000
#cICML
#index885
#id885
#%
#! model document citation learns identify hubs authorities set linked documents pages retrieved world wide web papers retrieved research paper archive unlike popular hits algorithm relies dubious statistical assumptions model provides probabilistic estimates clear semantics general identified authoritative documents correspond better human intuition

#* overview trec web track
#@Nick Craswell,David Hawking
#t2002
#cTREC
#index886
#id886
#%
#! trec web track moved away non web relevance ranking webspecific tasks million page crawl gov topic distillation task involved finding pages relevant characteristics make desirable inclusions distilled list key pages named page task variant years homepage finding task task particular page years task page need home page

#* effective site finding using link anchor information
#@Nick Craswell,David Hawking,S.E. Robertson
#t2001
#cSIGIR
#index887
#id887
#%
#! link based ranking methods described literature applied commercial web search engines according recent trec experiments better traditional content based methods conduct different type experiment task main entry point specific web site experiments ranking based link anchor text twice effective ranking based document content methods used bm formula obtained results using sets queries million document set set million document set site finding effectiveness begins explain search engines adopted link methods opens rich new area effectiveness improvement traditional methods fail

#* unification text link analysis
#@Brian D. Davison
#t2003
#cSIGIR
#index888
#id888
#%
#! paper presents simple profound idea thinking relationships terms documents generate richer representation encompasses aspects web link analysis text analysis techniques information retrieval paper shows path unified representation demonstrates use eigenvector calculations web link analysis stepping simple example

#* citation analysis tool journal evaluation
#@Eugene Garfield
#t1972
#cScience
#index889
#id889
#%
#! communications network journals play paramount role exchange scientific technical information little understood periodically gross gross published study references years issues journal american chemical socie pieces network illuminated work bradford allen gross woodford hooker henkle fussier brown map journal network whok date studies network interrelation components limited number journak areas scientific study periods time authors able consider shortcomings lack purpose insight energy investigators practical difficulty compiling manipulating manually enormous necessary data

#* graph theory practice
#@B Hayes
#t2000
#cAmerican Scientist
#index890
#id890
#%
#! diameter world wide web answer miles web truly world wide according albert lszl barabsi reka albert hawoong jeong notre dame university diameter web diameter question geometric distance concept comes branch mathematics called graph theory web place place clicking hypertext links makes sense define distance counting steps links question select web pages random links separate average million pages web room wander long paths barabsi et al know going just clicks mouse barabsi calculation reflects interesting shift style technology graph theory just years ago unusual apply graph theoretical methods enormous structure world wide web course just years ago web didn exist large netlike objects invite graph theoretical analysis time speak graph theory graph practice graph engineering

#* new status index derived sociometric analysis
#@L. Katz
#t1953
#cPsychometrika
#index891
#id891
#%
#! purpose evaluating status manner free deficiencies popularity contest procedures paper presents new method computation takes accountwho chooses ashow choose necessary introduce connection concept attenuation influence transmitted intermediaries

#* object level ranking bringing order web objects
#@Zaiqing Nie,Yuanzhi Zhang,Ji-Rong Wen,Wei-Ying Ma
#t2005
#cWWW
#index892
#id892
#%893,894,857,895,896,897,898,175,899,900,873
#! contrast current web search methods essentially document level ranking retrieval exploring new paradigm enable web search object level collect web information objects relevant specic application domain rank objects terms relevance popularity answer user queries traditional pagerank model longer valid object popularity calculation existence heterogeneous relationships objects paper introduces poprank domain independent object level link analysis model rank objects specic domain specically assign popularity propagation factor type object relationship study different popularity propagation factors heterogeneous relationships affect popularity ranking propose ecient approaches automatically decide factors experiments using million cs papers experimental results poprank achieve signicantly better ranking results naively applying pagerank object graph

#* wrapper generation semi structured internet sources
#@Naveen Ashish,Craig A. Knoblock
#t1997
#cSIGMOD Record
#index893
#id893
#%
#! current explosion information world wide web www wealth information different subjects available line numerous sources contain information classified semi structured present way access information browsing individual pages query web documents database like fashion based underlying structure provide database like querying semi structured www sources building wrappers sources present approach semi automatically generating wrappers key idea exploit formatting information pages source hypothesize underlying structure page structure generates wrapper facilitates querying source possibly integrating sources demonstrate ease able build wrappers number internet sources different domains using implemented wrapper generation toolkit

#* authority based keyword queries databases using objectrank
#@Andrey Balmin,Vagelis Hristidis,Yannis Papakonstantinou
#t2004
#cVery Large Data Bases
#index894
#id894
#%
#! objectrank applies authority based ranking keyword search databases modeled labeled graphs conceptually authority originates nodes objects containing keywords flows objects according semantic connections node ranked according authority respect particular keywords adjust weight global importance weight keyword query importance result actually containing keywords versus

#* block level link analysis
#@Deng Cai,Xiaofei He,Ji-Rong Wen,Wei-Ying Ma
#t2004
#cSIGIR
#index895
#id895
#%
#! link analysis shown great potential improving formance web search pagerank hits popular algorithms existing link analysis algorithms treat web page single node web graph cases web page contains multiple semantics web page considered atomic node paper web page partitioned blocks using vision based page segmentation algorithm extracting page block block page relationships link structure page layout analysis construct semantic graph www node exactly represents single semantic topic graph better semantic structure web based block level link analysis proposed new algorithms block level pagerank block level hits performances study extensively using web data

#* xrank ranked keyword search xml documents
#@Lin Guo,Feng Shao,Chavdar Botev,Jayavel Shanmugasundaram
#t2003
#cSIGMOD Conference
#index896
#id896
#%
#! consider problem efficiently producing ranked results keyword search queries hyperlinked xml documents evaluating keyword search queries hierarchical xml documents opposed conceptually flat html documents introduces new challenges xml keyword search queries return entire documents return deeply nested xml elements contain desired keywords second nested structure xml implies notion ranking

#* discovering complex matchings web query interfaces correlation mining approach
#@Bin He,Kevin Chen-Chuan Chang,Jiawei Han
#t2004
#cKDD
#index897
#id897
#%
#! enable information integration schema matching critical step discovering semantic correspondences attributes heterogeneous sources complex matchings common far complex search space existing techniques focus simple matchings tackle challenge paper takes conceptually novel approach viewing schema matching correlation mining task matching web query interfaces integrate myriad databases internet deep web query interfaces generally form complex matchings attribute groups corresponds books domain observe occurrences patterns query interfaces reveal complex semantic relationships grouping attributes tend present query interfaces positively correlated contrast synonym attributes negatively correlated rarely occur insight enables discover complex matchings correlation mining approach particular develop dcm framework consists data preparation dual mining positive negative correlations finally matching selection unlike previous correlation mining algorithms mainly focus finding strong positive correlations algorithm cares positive negative correlations especially subtlety negative correlations special importance schema matching leads introduction new correlation measure measure distinct proposed previous work evaluate approach extensively results good accuracy discovering complex matchings

#* optimization simulated annealing
#@S. Kirkpatrick,C. D. Gelatt,M. P. Vecchi
#t1983
#cScience
#index898
#id898
#%
#! deep useful connection statistical mechanics behavior systems degrees freedom thermal equilibrium finite temperature multivariate combinatorial optimization finding minimum given function depending parameters detailed analogy annealing solids provides framework optimization properties large complex systems connection statistical mechanics exposes new information provides unfamiliar perspective traditional optimization problems methods

#* wrapper induction information extraction
#@Nicholas Kushmerick,Daniel S. Weld,Robert B. Doorenbos
#t1997
#cIJCAI (1)
#index899
#id899
#%
#! internet presents numerous sources useful information telephone directories product catalogs stock quotes weather forecasts recently systems built automatically gather manipulate information user behalf resources usually formatted use people relevant content embedded html pages extracting content difficult wrappers used purpose wrapper procedure

#* mining data records web pages
#@Bing Liu,Robert L. Grossman,Yanhong Zhai
#t2003
#cKDD
#index900
#id900
#%
#! large information web contained regularly structured objects data records data records important present essential information host pages lists products services useful data records order extract information provide value added services existing approaches solving problem mainly include manual approach supervised learning automatic

#* automatic personalization based web usage mining
#@Bamshad Mobasher,Robert Cooley,Jaideep Srivastava
#t2000
#cCommun. ACM
#index901
#id901
#%
#! ease speed business transactions carried web key driving force rapid growth commerce ability track user browsing behavior individual mouse clicks brought vendor end customer closer possible vendors personalize product messages individual customers massive scale phenomenon referred mass customization course type

#* personal webwatcher design implementation
#@Dunja Mladenic
#t1996
#cTechnical report
#index902
#id902
#%
#! growing availability information sources especially non homogeneous distributed sources like world wide web growing tools help making good quick selection information interested

#* syskill webert identifying interesting web sites
#@M Pazzani,J Muramatsu,D Billsus
#t1996
#cAAAI
#index903
#id903
#%
#! syskill webert software agent learns rate pages world wide web www deciding pages user user rates explored pages point scale syskill webert learns user profile analyzing information page user profile used ways used suggest links user interested exploring second used construct lycos query pages user compare different algorithms machine learning information retrieval task naive bayesian classifier offers advantages learning algorithms task furthermore initial portion web page sufficient making predictions interestingness substantially reducing network transmission required make predictions

#* feature weighting instance selection collaborative filtering
#@Kai Yu,Zhong Wen,Xiaowei Xu,Martin Ester
#t2001
#cKnowledge and Information Systems
#index904
#id904
#%
#! collaborative filtering uses database consumers preferences make personal product recommendations achieving widespread success commerce nowadays paper present feature weighting methods improve accuracy collaborative filtering algorithms furthermore propose reduce training data set selecting highly relevant instances evaluate various methods known eachmovie data set experimental results mutual information achieves largest accuracy gain feature weighting methods interesting fact data reduction method achieves improvement accuracy speeding collaborative filtering algorithm factor

#* instance selection techniques memory based collaborative filtering
#@Kai Yu,Xiaowei Xu,Jianjua Tao,Martin Ester,Hans-Peter Kriegel
#t2002
#cSDM
#index905
#id905
#%
#! collaborative filtering cf important data mining technique make personalized recommendations books web pages movies popular algorithm memory based collaborative filtering predicts users preference based similarity users instances database tremendous growth users large number products memory based cf algorithms results problem deciding right instances use prediction order reduce executive cost excessive storage possibly improve generalization accuracy avoiding noise overfitting paper focus work typical user preference database contains missing values propose novel instance reduction techniques called turf turf preprocessing step improve efficiency accuracy memory based cf algorithm key idea generate prediction carefully selected set relevant instances evaluate techniques known eachmovie data set experiments showed proposed algorithms just dramatically speed prediction improved accuracy

#* probabilistic analysis rocchio algorithm tfidf text categorization
#@Thorsten Joachims
#t1997
#cICML
#index906
#id906
#%
#! rocchio relevance feedback algorithm popular widely applied learning methods information retrieval probabilistic analysis algorithm presented text categorization framework analysis gives theoretical insight heuristics used rocchio algorithm particularly word weighting scheme similarity metric suggests improvements lead probabilistic variant rocchio classifier rocchio classifier probabilistic variant naive bayes classifier compared text categorization tasks results probabilistic algorithms preferable heuristic rocchio classifier founded achieve better performance

#* knowledge discovery large spatial databases focusing techniques efficient class identification
#@Martin Ester,Hans-Peter Kriegel,Xiaowei Xu
#t1995
#cSSD
#index907
#id907
#%
#! number size spatial databases rapidly growing large data obtained satellite images ray crystallography scientific equipment automated knowledge discovery important spatial databases far methods knowledge discovery databases kdd based relational database systems paper address task class identification spatial databases using clustering techniques special emphasis integration discovery methods db interface crucial efficiency kdd large databases key integration use known spatial access method tree focusing component kdd determines parts database relevant knowledge discovery task present strategies focusing selecting representatives spatial database focusing relevant clusters retrieving objects given cluster applied proposed techniques real data large protein database used predicting protein protein docking performance evaluation database indicates clustering large spatial databases performed efficiently effectively research funded german minister research technology bmft grant ib authors responsible contents paper

