A knowledge-based approach for pseudo-relevance feedback by exploiting semantic relevance

Junmei Wang, Jing Zeng, Jimmy X. Huang, Luyun Wang, Jiajia Wang

Abstract: For most existing query expansion technologies in pseudo-relevance feedback approaches, expansion terms are mainly selected in terms of the term frequency, the cooccurrence frequency, term proximity and so on. However, terms that are close to the query term may not be semantically related to query. In this paper, we propose a new knowledge-based query expansion approach by exploiting semantic relevance. To avoid selecting terms that have high term frequency but low semantic relevance, we expand a semantic set for the original query by means of ConceptNet and use the importance of the terms in the feedback documents to filter the semantic description set. Then, we propose a method for calculating semantic relevance between the expansion term and query. Besides, to test the effectiveness, the proposed knowledge-based query expansion approach is applied to Rocchio and RM3 models. Finally, through evaluations and comparisons on TREC data sets, results display that the improved models are superior to the well-known baseline models and comparable to the competitive pseudo-relevance feedback models.

External IDs:dblp:journals/kais/WangZHWW25