When selecting expansion terms from both clicked documents and queries in log,path consistent has the influence to the prior choice of expansion terms in each resource.Because current model assigns the same expansion term resource of different queries by the same path consistent value,all queries may prior select the same expansion term resource.However,it can be observed from log that different queries are suitable for selecting expansion terms from different expansion term resources.Therefore,this paper proposed an expanded Path-Constrained Random Walks model.The model firstly evaluates relevance between the query and expansion terms from different resources by a unified judgment,and assigns different expansion term resources of one query with different path consistent values.That allows different queries can prior select different resources of expansion term by their own condition.Experiment results show that performance of expansion query generated by this paper is better and more related to current user intent.

User intent modeling is a hot point in researches of query expansion and query recommendation.But the large amount of noise in search log have great negative impact on the construction of user intent model.By observing the log it can be found that tentative click of the user is one of the main causations of irrelevant feedbacks.To solve the problem,this paper proposed the user intent model combined with studying process.It modeled the tentative clicks,and then identified and filtered those tentative clicks.Test results show that the model can effectively filter noises coming from tentative clicks when log owns high level noisy.After applying the model to query recommendation it can improve the precision of similarity computation among queries,and increases the precision of query recommendation obviously.

传统的成果地质资料查询服务系统是基于数据库查询设计的,以项目档案的标题和内容摘要为主进行查询,存在借阅者查找资料时间长,检索精度不高,甚至漏检部分资料的问题,为此,本文重新设计并实现了成果地质资料检索系统.通过对汇交的项目文档进行处理,建立索引,利用全文检索技术提升搜索结果的召回率;基于本体和搜索日志,使用查询扩展技术提升搜索结果的准确率.经资料馆应用证明,该系统节约了借阅者寻找资料的时间,提高了检索性能,增强了资料馆的服务能力.

There are some users that own some records in the search log,but it is not enough to provide the users with personalized query recommendation service.For these users,it can improve their satisfaction with recommendations results if targeted recommendation services can be provided.In order to solve the problem,this paper proposes a differentiated query recommendation method for different search background.Users are divided into different groups according to user search behavior.Different recommendation service is provided for different groups to realize differentiated recommendation service between group users.Experimental results on real datasets show that this method can reduce the risk of recommendation failure and improve the satisfaction of users for recommendation.

基于日志的搜索引擎查询推荐方法大多从搜索行为流行性考虑进行查询推荐,这类推荐方法忽略了个体用户背景不同带来的搜索需求的差异,对于一些在搜索引擎日志中有较多记录的用户来讲,难以满足他们更高质量的搜索需求.然而,此类用户的搜索记录又不足以支持系统提供个性化的推荐服务.所以,针对在搜索引擎日志中有较多记录的用户,本文提出一种折衷的解决方案,即流行性与相似性结合的查询推荐策略,提供一种近似于个性化的推荐服务.本文采用社团发现的方法发现搜索日志中的密集行为,从而构建典型用户行为模型代表用户搜索背景,并引入模块度度量典型用户行为的强度,最后结合流行性、群体用户对推荐词的支持程度、用户对用户群体的归属程度以及典型用户行为的强度来产生推荐词集合.本文最后,在真实数据集的实验结果证明了本策略能够在控制推荐失败风险的同时提高用户对推荐结果的满意度.

基于语义资料和局部分析的混合式查询扩展可以同时提供具有语义相关性和时效性的扩展结果,但如何有效地混合不同相似度度量指标是尚未解决的问题.提出了一种基于Copulas框架的混合式查询扩展方法,在统一框架内实现了不同类型相似度度量指标的合并.该方法基于语义分析及词语共现分析方法,分别计算扩展词与用户查询词的语义及统计相似概率,进而在Copulas框架下融合扩展词集,选取最高质量的扩展词形成查询扩展.实验结果表明,该方法充分利用了语义及词语共现分析查询扩展方法的优点,有效地弥补了两者的不足,提高了搜索结果的查准率,具有更优的搜索性能.

For the question that query expansion words extracted by query expansion method based on search log are seriously influenced by the popularity of the words,which leads to a period covered with limited knowledge,and cause the missing part of the search results to satisfy users' information needs,this paper studies a combined query expansion method for balanced epidemic and similarity.On one hand,this method clusters search logs,matching search queries and generate log based query expansions.Meanwhile,this method generates user query expansion word classification set using ontologies.Based on the classification set,the semantic coverage of the log based expansion set is calculated,and for expansion sets with low semantic coverage,ontology based expansion results are merged into log based expansion set using theory of evidence.The experimental results show that the method improves the retrieval performance of search engine,and it can meet the needs of the users to search information,which is able to improve the satisfaction of users.

当搜索日志存在查询词稀疏性和时效性问题时,基于搜索日志的查询扩展方法无法保证查询扩展的质量,无法满足具有时效性查询请求的需求.提出了基于搜索日志与局部上下文的查询扩展方法,通过发掘搜索日志中用户查询词和相关文档的关联关系,抽取查询扩展词,并用局部上下文从相关文档集中提取出频率较大,同时具有时效性的查询扩展词.通过对查询扩展词的查询性能和时效性的计算,该方法为原始查询补充更高质量的查询扩展词.实验结果表明,该方法能够有效地提升准确率和召回率,使原始查询获得更好的查询性能.

During selection of expansion terms from both clicked documents and queries in log, current methods never consider the large amount of irrelevant feedbacks in records of new queries, which leads to the inconsistency between users' intents and selected expansion terms. To solve this problem, this paper integrated the ranking model with the recognition of irrelevant document content according to users' study processes, so it can reduce the weights of irrelevant expansion terms. The experimental results show that this method can measure relevancy between query and document terms better, especially in the situation when records have more irrelevant feedbacks, and can select expansion terms accorded with users' intent better.

In order to make search results better fit users’ current search interest, this paper proposes an algorithm of repeated results re-ranking using a model of polysemy. The algorithm considers the characteristics of the keywords to improve the rank of repeated results. Based on the analysis of polysemy of the keywords, we propose a polysemous model of concept lattice, then we combine with the user interest model to change the rank of repeated results in a search session. The method of this paper considers the impact of polysemy of keywords which may improve the ranking. The experimental results show that the process based on the polysemy of the keyword can reduce the length of the search session, especially when the keywords have multiple meanings.

Query expansion methods based on search logs could improve the quality of search results to some extends. But when the search logs are sparse, this kind of query expansion methods will have poor quality of search results and are unable to meet the user's search request, etc. This paper presents the search log sparseness oriented query extension method. By introducing the determination rule of data sparseness, this method selects expansion terms with high performance from the expansion terms given by local context based methods to go over the disadvantages of search log based method with sparse data sets, providing expansion terms with higher quality for the user's initial queries. The experimental results show that, this method improves the accuracy and recall of the search results, improving the quality of search results.

Research on the relationship of data resource has become a research focus in data space area. Current data relationship researches mainly concentrate on the self attributes of data, however, the effect caused by users is neglected. As a consequence, user attention could hardly be expressed during the management, organization and application of data. User behavior based data relationship acquisition method was proposed to decide whether the user behavior is intentional by R/S analyzing of user history behavior. The user behavior sequence was excavated based on this to get user behavior pattern, and finally data relationship of data contained by the pattern in form of iTrail was built. The experiments proved that the method had a good effect.

In the area of Deep Web data integration, it is a key issue to efficiently and accurately build result output pattern according to uses' query. However, it is the crucial problem to improve efficiency and precision of result output pattern generating that resolving repeat pattern matching and data heterogeneity. This paper proposes the approach of Deep Web data integration oriented result output pattern building. On the basis of the conflicts and conflict resolving rules between any two data sources, the approach gives the rules for conflict integration over multi-sources. According to the result patterns and conflicts between result patterns, this paper gives the mechanism and algorithm of result output pattern building which can effectively resolve the problems of repeat pattern matching and data heterogeneity and lay a good foundation for Deep Web data merging.

Social tagging is one of the most important ways to organize and index online resources. Recommendation in social tagging systems, e.g. tag recommendation, item recommendation and user recommendation, is used to improve the quality of tags and to ease the tagging or searching process. Existing works usually provide recommendations by analyzing relation information in social tagging systems, suffering a lot from the over sparse problem. These approaches ignore information contained in the content of resources, which we believe should be considered to improve recommendation quality and to deal with the over sparse problem. In this paper we propose a recommendation approach for social tagging systems that combines content and relation analysis in a single model. By modeling the generating process of social tagging systems in a latent Dirichlet allocation approach, we build a fully generative model for social tagging, leverage it to estimate the relation between users, tags and resources and achieve tag, item and user recommendation tasks. The model is evaluated using a CiteULike data snapshot, and results show improvements in metrics for various recommendation tasks.

In a social tagging system,a user's tagging habits,including choosing which resource to tag and using which tag to annotate a resource,are affected by one's own autonomy.Available personalized rag recommendation methods lack the ability to model such autonomy information,and limit the performance of these methods.This paper proposed a latent Dirichlet allocation like probabilistic approach,which modeled user autonomy information such as one's preferences on tag and resource use,to provide autonomy oriented personalized tag recommendations.The parameters of the proposed method were estimated following a Gibbs sampling approach,which allowed a quick calculation of the values.Experiment results showed that the proposed approach can provide personalized tag recommendations with higher quality.

The critical reality in developing Ontology driven applications is that there may not be an ontology available at hand. The idea of automatically extracting ontologies from Wiki and Semantic Wiki has been proposed, but Wiki and Semantic Wiki themselves are facing several problems and may affect the ontology building process. In this paper we present our personal knowledge management tool APKM, and describe how to build ontologies from personal knowledge bases created by netizens with APKM. An iterative approach that build ontology from personal knowledge bases is then proposed. 1553-9105/ Copyright © 2009 Binary Information Press.

The query interface integration is the key technology in realization of Deep Web information integration. However the bottleneck problem is the solution of differences and conflicts of query interfaces. TheThe most existed query interface integration work mainly focused on the conflict of attribute labels and attribute value, but there are many kinds of conflicts between interfaces because of the special nature of interfaces, such as attribute set structure conflict, attribute requirement conflict and so on. Ignore these conflicts will lead to incomplete and inaccurate integrated interface. Based on the query interface features, this paper presents the formal definition of interface, gives the framework for classification of the conflicts in query interface integration, and for each kind of conflict there is a corresponding solution strategy that lays a good foundation for interface integration.