Font Size: a A A

Research And Application On Expansion Term Ranking Model For Query Understanding

Posted on:2019-03-28Degree:DoctorType:Dissertation
Country:ChinaCandidate:B XuFull Text:PDF
GTID:1368330548484759Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
In recent years,information retrieval has attracted extensive attention of researchers and practitioners with the continuously increasing amounts of data on the Internet.Information retrieval methods have been widely applied in various vertical information retrieval tasks.Information retrieval aims to retrieve relevant documents or websites based on given queries for producing satisfactory ranking lists of results.Queries submitted by search engine users always contain a few keywords.The keywords may be insufficient to describe users'information needs,thus bringing ambiguities and uncertainties for interpreting query intents.Therefore,how to accurately understand users' queries becomes more and more crucial for improving retrieval performance.Query understanding becomes one of the most important issues in optimizing information retrieval systems.To accurately understand users' queries and improve retrieval performance,this dissertation carries out research from three aspects as follows:1.In order to solve the problem of term selection in query expansion,this dissertation proposes an expansion term ranking model based on pseudo relevance feedback.The model takes learning to rank as the core technology to select expansion terms for fully meeting users' information needs.The selected terms are used to enrich users' queries and improve the quality of expanded queries.In model construction,multiple optimization strategies are adopted in terms of candidate expansion term selection,term relevance labeling,term feature extraction and ranking model construction,respectively.Extensive experiments are conducted based on three standard TREC datasets.Experimental results show that the proposed model can effectively improve retrieval performance in general fields.2.In order to better understand queries in biomedical literature retrieval,this dissertation proposes a biomedical term ranking model.The model is based on the proposed term ranking model for pseudo relevance feedback,and incorporates biomedical knowledge to optimize the selection and feature extraction of expansion terms.Candidate expansion terms are labeled based on the topic terms of corresponding queries.In model construction,a group-based term ranking model is proposed based on group-wise learning to rank to optimize the sampling space of biomedical term ranking.Experimental results on two standard datasets from TREC Genomics tracks show that the proposed model is effective in improving the performance of biomedical literature retrieval.3.In order to better understand queries in code snippet-oriented information retrieval,this dissertation proposes a code term ranking model.The model is based on the proposed term ranking model for pseudo relevance feedback,and integrates information of code snippets and code files into the optimization of candidate expansion term selection,term feature extraction and term labeling strategy,respectively.In model construction,a query-level autoencoder based term ranking model is proposed to optimize the feature space of term samples for improving term ranking precision.Experimental results on existing code snippet datasets show that the proposed model can effectively improve the performance of code snippet retrieval.Research in this dissertation focuses on understanding queries in information retrieval by constructing term ranking models,and applies the proposed models to two vertical retrieval tasks.The proposed models improve the performance of term selection in query expansion for accurately understanding users' query intents.Furthermore,the proposed models can be generalized and applied to other domain-specific retrieval tasks.
Keywords/Search Tags:Query expansion, Learning to rank, Literature retrieval, Code retrieval, Pseudo relevance feedback
PDF Full Text Request
Related items