Effectiveness of Query Expansion based on Clustering of Pseudo-Feedback Documents with K-NN Algorithm

Query expansion is one of the effective approaches for improving effectiveness of information retrieval. Pseudo-relevance feedback (PRF) supposes that top-ranked documents of retrieval from primary retrieved results are relevant to the query and selects some relevant terms from top-ranked documents for expansion. Existence of noisy documents in the top ranked documents drives researchers toward inventing approaches for selecting best documents as source for selecting expansion terms. The selection of the best documents for extraction of relevant terms for expansion is the most important issue in query expansion. In this paper, we propose clustering of pseudo feedback documents (CPRF), selected from primary results, based on cosine similarity to place the most similar documents beside each other. Some clusters are selected as feedback ones based on their inner content and some top ranked documents of them are selected as feedback documents. A combined document is constructed from selected documents and terms of combined document are ranked by term frequency-inverse document frequency (TF-IDF) schema. High ranked terms are selected for query expansion. Experimental results over MED collection shows postulated approach overcome pseudo relevance feedback approach respect to average mean accuracy (MAP).