Abstract

Abstract: The application of natural language processing tools and techniques to information retrieval tasks has long since been identified as potentially useful for the quality of information retrieval. Traditionally, IR has been based on matching words or terms in a query with words or terms in a document. In this paper we introduce an approach to IR based on computing a semantic distance measurement between concepts or words and using this word distance to compute a similarity between a query and a document. Two such semantic distance measures are presented in this paper and both are benchmarked on queries and documents from the TREC collection. Although our results in terms of precision and recall are disappointing, we rationalise this in terms of our experimental setup and our results show promise for future work in this area. 1

Citations