Abstract

In most information retrieval models, keyword frequency plays a major role. However, these models are less context dependent, the main disadvantage is that they don't have a deeper understanding of the documents. The traditional relevance feedback techniques assume that terms in documents are independent; they do not consider the term dependencies and the semantic relations among documents. In this paper, we propose a Word Co-occurrence Matrix based Method for relevance feedback. Firstly, the definition of word co-occurrence matrix is given. Unlike other studies about word association, we consider the influence of the inter word distance and co-windows ratio. So we think that this matrix can simply represents the document semantic relations and then we calculate the similarity between documents according to this matrix. In the feedback process, this similarity score will be combined with the initial score to improve the retrieval effectiveness. Experiments with TREC dataset demonstrate the effectiveness of this method.

Paper Details