A Clustering Approach to Improving Pseudo-Relevance Feedback: Improving Retrieval Effetiveness by Removing Noisy Documents

摘要

Pseudo relevance feedback is an effective technique for improving retrieval results, which assumes a small number of top-ranked documents in the initial retrieval results are relevant and selects from these documents related terms to the query to improve the query representation through query expansion. However, these documents are often a mixture of relevant and irrelevant documents. The relevance feedback is quite effective and performs significantly better than pseudo-relevance feedback, which needs the user explicitly provides information on relevant documents to a query. This paper makes a case for the use of query-specific density clustering in IR on the grounds of improved retrieval effectiveness in a fully automatic manner and without relevance information provided by human and the experimental results show that significant improvements can be obtained on several collections when our new model FWN (Feedback Without Noise) is used.