Text Retrieval Based on Semantic Relationship

Expansion of query keywords based on semantic relationship is an effective approach to improve the performance of text retrieval. In this paper, a novel approach for text retrieval is presented. The principle of the approach is to construct a integrated semantic tree, and select candidate keywords from the tree. On the tree, all nodes are weighted based on synonymy, hypernymy, and Mutual Information. The weights of nodes will be used to supplement tfidf values in computing the similarity between query and documents. Experimental results demonstrate about 14.6% precision and 13.7% prec@20 improvement over the traditional tfidf-based method.