The Research Of Clustering And Local Co-occurrence For Relevance Feedback Method

The rapid development of Internet has led to the increasing amount of informationon the Internet. It has become a hot issue to find the information users need from theInternet. In this context, information retrieval technology developed. For the wordmismatch problem between document and query in information retrieval, a large numberof studies have shown that relevance feedback based on query expansion can be a goodsolution to this problem and improve the information retrieval system performance.Most typical query expansion methods assume the top-retrieved documents arerelevant and use these documents to expand terms. These feedback documents can,however, contain a great deal of noise, and the new query may drift away from the querytopic, and lead to bad information retrieval performance. As to this problem, we present aclustering method to select better feedback documents based on relevance model. Themain idea is to use document clusters to find relevant documents for the initial retrievalset, and to drop non-relevant documents.Meanwhile most typical query expansion methods expand terms based on term’sweight in feedback documents. These methods didn’t consider the relationship betweenterm in feedback documents and the query topic, and this may affect the quality ofexpanded terms, and result in bad performance in information retrieval. A localco-occurrence method can solve this problem.A method based on both clustering and local co-occurrence can better improve theinformation retrieval system performance.Experimental results show significant improvements over the methods proposed.This result indicates that the proposed methods are effective for relevance feedback.