Abstract

Query expansion is one of the key important problems in distributed information retrieval. Generally, resources in distributed location are different in topic, and also are different from initial queries user expressed. Thus, these initial queries commonly can not be used to describe user's retrieval intention precisely, which are need to be modified according to query-specific retrieval results. In this paper, we introduce a cluster-based collection correlative model for query expansion to avoid interest drift in distributed information retrieval. Firstly, Mixture model is used to get pseudo feedback documents in query-specific collections, and then the cluster and global information are introduced to construct local resource correlative query expansion model, which is used to get query expansion terms. Finally, expanded queries, which are closer to user's topic, are generated by adding these specific terms to initial query. The experimental result shows that our approach consistently improves retrieval performance over initial query, global and local-based expansion strategies.

Paper Details