Rocchio's Model Based on Vector Space Basis Change for Pseudo Relevance Feedback

Rocchio's relevance feedback model is a classic query expansion method and it has been shown to be effective in boosting information retrieval performance. The main problem with this method is that the relevant and the irrelevant documents overlap in the vector space because they often share same terms (at least the terms of the query). With respect to the initial vector space basis (index terms), it is difficult to select terms that separate relevant and irrelevant documents. The Vector Space Basis Change is used to separate relevant and irrelevant documents without any modification on the query term weights. In this paper, first, we study how to incorporate Vector Space Basis Change into the Rocchio's model. Second, we propose Rocchio's models based on Vector Space Basis Change, called VSBCRoc models. Experimental results on a TREC collection show that our proposed models are effective.