Query Expansion based on Associated Semantic Space

Abstract

Expansion of query keywords based on semantic relations is an effective approach to improve the performance of information retrieval. Traditional methods of query expansion did not adequately make use of semantic relations between query keywords. In this paper, a novel approach for query expansion is presented. The main idea of the approach is to construct a ‘Tree of Associational Semantics Model’ and select candidate keywords from the tree. In the first step, a group of initial semantic trees for original keywords are constructed based on WordNet thesaurus. Secondly, noise nodes on the trees are removed by calculating the similarities between words. The pruned trees are subsequently assembled into a big integrated tree, i.e. Tree of Associational Semantics Model, by expanding the trees upward until finding a common root. Finally , the nodes on the integrated tree are filtered and supplemented based on Mutual Information. All words selected from the tree are assigned semantic weight s which are used in computing similarity between the query and documents in internet. In addition, the distributional situation of query keywords in documents is also considered in document retrieval. Experimental results demonstrate about 14.6% precision and 13.7% prec@20 improvement over the traditional tfidf -based method.