The Impact of Term Statistical Relationships on Rocchio’s Model Parameters For Pseudo Relevance Feedback

Abstract

Query Expansion using the Pseudo Relevance Feedback based on rocchio’s model is a popular technique for reformulating the original user’s query. This latter, assumes that most frequent terms in the returned documents are useful to ameliorate the original query and therefore to improve search results which is a challenge today with the proliferation of textual data on the web. In this study, we re-examine this assumption and show that it does not hold in reality. Indeed, many expansion terms identified are unrelated to the query and reduce the performance of the retrieval system. In this paper, we present our method to revisit the rocchio’s model parameters in order to take into account the relationships between terms which are defined by our proposed statistical method based on least square optimization. The evaluation process was performed on CLEF-eHealth-2014 database which is composed of about one million medical and english documents and 50 professional and medical queries. Experimental results show that our proposed method is effective.