Pattern-based Topic Modelling for Query Expansion

Gao, Y., Xu, Y. and Li, Y.

    One big problem with information retrieval (IR) is that the size of queries is usually short and the key- words in a query are very often ambiguous or inconsistent. Automatic query expansion is a widely recognized technique which is effective to deal with this problem. However, many query expansions methods require extra information such as explicit relevance feedback from users or pseudo relevance feed- back from retrieval results. In this paper, we propose an unsupervised query expansion method, called Topical Query Expansion (TQE), which does not require extra information. The proposed TQE method expands a given query based on the topical patterns which can create links among those more associated and semantic words in each topic. This model also discovers related topics that are related to the original query. Based on the expanded terms and related topics, we propose to rank the document relevance with different ranking strategies. We conduct experiments on popularly used datasets, TREC datasets, to evaluate the proposed methods. The results demonstrate outstanding results against several state-of-the- art models.
Cite as: Gao, Y., Xu, Y. and Li, Y. (2014). Pattern-based Topic Modelling for Query Expansion. In Proc. Twelfth Australasian Data Mining Conference (AusDM14) Brisbane, Australia. CRPIT, 158. Li, X., Liu, L., Ong, K.L. and Zhao, Y. Eds., ACS. 165-173
pdf (from crpit.com) pdf (local if available) BibTeX EndNote GS