Query expansion using the clustering of pseudo relevant documents with query sensitive similarity

Query expansion as one of query adaptation approaches, improves retrieval effectiveness of information retrieval. Pseudo-relevance feedback (PRF) is a query expansion approach that supposes top-ranked documents are relevant to the query concept, and selects expansion terms from top-ranked documents. However, Existing of irrelevant document in top-ranked documents is possible. Many approaches have been proposed for selecting relevant documents and ignoring irrelevant ones, which use clustering or classification of documents. Important issue in query expansion approaches is using relevant documents for selecting expansion terms. In this paper, we propose clustering of pseudo-relevant documents based on query sensitive similarity, which is efficient for placing similar documents together. Query sensitive similarity obtained good results in document retrieval rather than term-based similarity, is the reason for using in this paper. Clusters are ranked based on inner similarity, and some top ranked ones are selected for query expansion. Then, we extract expansion terms from documents of selected clusters based on Term Frequency- Inverse document frequency (TF-IDF) scoring function. Conducted experiments over Medicine dataset (MED) shows that retrieval results for expanded queries with selected documents from clusters is better than basic retrieval (VSM) and Pseudo-relevance feedback. In addition, the effectiveness of retrieval is raised.