A new query expansion approach using collocation relationships in language models for information retrieval

Language Modeling (LM) has become very popular in Information Retrieval (IR) thanks to its sound theoretical basis and good empirical success. It has proven to be capable of integrating term relations and query expansions. In some recent research in order to relax the independence assumption, dependency models have been proposed to incorporate term relations into LM for query expansion. While most of them depend on either WordNet or co-occurrence relations, no explicitly term syntactic or semantic relationships are used. Different from these methods, in this paper, we select term collocation relationships with both linguistics and statistical significance for query expansion. Another difference is that term collocation relationships are used to expand query model instead of document model in this study. Local Relevance Feedback (LCA) has already been shown to be an effective method in improving retrieval performance. We combine important term collocation relationships with LCA documents to further improve performance. Our experiments on three TREC collections show that this new type of collocation relations performs much better than traditional term relations. © 2009 Binary Information Press.