Searches in patent collections to determine if a given patent application has related prior art patents is non-trivial and often requires extensive manpower. When time is constrained, an automatically generated, ranked list of prior art patents associated with a given patent application decreases search costs and improves search efficiency. One may view the discovery of this prior art patent set as a problem of finding patents ‘related’ to the patent application. To accomplish this, we examine whether semantic relations between patent classification codes can aid in the recognition of related prior art patents. We explore similarity measures for hierarchically ordered patent classes and subclasses for this purpose. Next, we examine various patent feature-weighting schemes to achieve the best similarities between our patent applications and related prior art patents. Finally, we provide a method and demonstrate that patent prior art searches can successfully be used as an aid in patent ranking.

Information overload is a well-known problem facing biomedical professionals. MEDLINE, the biomedical bibliographic database, adds hundreds of articles daily to the millions already in its collection. This overload is exacerbated by the lack of relevance-based ranking for search results, as well as disparate levels of search skill and domain experience of professionals using systems designed to search MEDLINE. We propose to address these problems through learning ranking functions from user relevance feedback. Simple active learning techniques can be used to learn ranking functions using a fraction of the available data, with performance approaching that of functions learned using all available data. Furthermore, ranking functions learned using metadata features from the Medical Subject Heading (MeSH) terms associated with MEDLINE citations greatly outperform functions learned using textual features. An in-depth investigation is made into the effect of a number of variables in the ranking round, while further investigation is made into peripheral issues such as users providing inconsistent data.

Patent classification systems are used to help scrutinize patent applications for possible violations of the novelty and non-obviousness/inventive steps of a patentability test. There are several different patent classification systems in use today, each with a different underlying philosophy and approach. We compare the two most widely-used patent classification systems -- the IPC and USPC -- and examine their ability to help re-rank patents based on similarity. We observed a significant improvement in MAP, Recall@100, and nDCG when using these systems to re-rank our retrieved document set, demonstrating their overall utility in patent searches.

Searches on patents to determine prior art violations are often cumbersome and require extensive manpower to accomplish successfully. When time is constrained, an automatically generated list of candidate patents may decrease search costs and improve search efficiency. We examine whether semantic relations inferred from the pseudo-hierarchy of patent classifications can contribute to the recognition of related patents. We examine a similarity measure for hierarchically-ordered patent classes and subclasses and return a ranked list of candidate patents, using a similarity measure that has demonstrated its effectiveness when applied to WordNet ontologies. We then demonstrate that this ranked list of candidate patents allows us to better constrain the effort needed to examine for prior art violations on a target patent.

: The University of Iowa Team, participated in the blog track and the chemistry track of TREC-2009. This is our first year participating in the blog track as well as the chemistry track.

We indexed the entire TREC collection, opting to merge the document content and the title into a single field, using the Lucene StandardAnalyzer, which strips punctuation, but recognizes and retains elements such as e-mail addresses. The StandardAnalyzer stoplist was used for indexing. For our explorations, we converted topic fields into term vectors for querying the collection. For each topic, our system returned a ranked set of results with enough documents to match in quantity either those retrieved by a reference Boolean query executed on behalf of the TREC 2006 evaluators, or enough to reach a set cap on the number of documents returned, whichever was greater.

We have built web interfaces to a number of Natural Language Processing technologies. These interfaces allow students to experiment with different inputs and view corresponding output and inner workings of the systems. When possible, the interfaces also enable the student to modify the knowledge bases of the systems and view the resulting change in behavior. Such interfaces are important because they allow students without computer science background to learn by doing. Web interfaces also sidestep issues of platform dependency in software packages, available computer lab times, etc. We discuss our basic approach and lessons learned.

An information extraction system is designed to operate over a specific domain, and cannot be applied to new domains without being adapted if it is to perform well. We will investigate the problem of adapting information extraction systems to new domains by first defining the task of information extraction and giving an example of an information extraction system. We will then outline the modules comprising a typical system designed to solve the task and comment on their impact on domain adaptation. Previous work on adapting systems to new domains will be reviewed, along with work by the author. We will conclude with open problems and future areas of research in this field.

Tokenization in the bioscience domain is often difficult. New terms, technical terminology, and nonstandard orthography, all common in bioscience text, contribute to this difficulty. This paper will introduce the tasks of tokenization, normalization before introducing BAccHANT, a system built for bioscience text normalization. Casting tokenization / normalization as a problem of punctuation classification motivates using machine learning methods in the implementation of this system. The evaluation of BAccHANT's performance included error analysis of the system's performance inside and outside of named entities (NEs) from the GENIA corpus, which led to the creation of a normalization system trained solely on data from inside NEs, BAccHANT-N. Evaluation of this new system indicated that normalization systems trained on data inside NEs perform better than systems trained both inside and outside NEs, motivating a merging of tokenization and named entity tagging processes as opposed to the standard pipelining approach.

LEARNING TO RANK DOCUMENTS WITH SUPPORT VECTORMACHINES VIA ACTIVE LEARNINGbyRobert James ArensAn AbstractOf a thesis submitted in partial ful llment of therequirements for the Doctor of Philosophydegree in Computer Science in theGraduate College of TheUniversity of IowaDecember 2009Thesis Supervisor: Professor Alberto Segre