About

Although web search engines were highly powerful, they often failed to provide intuitively relevant results for many types of queries, particularly those that were vaguely formed in the user’s mind. Researchers argued that associations between terms in a search query could reveal the underlying information needs of users and should be considered in search processes.

A multi-faceted approach was proposed to detect and exploit such associations. The CORDER method measured the association strength between query terms, identifying queries with low association strength as vague queries. For these queries, WordNet was used to find related terms and compose extended queries, with a particular focus on the role of least common subsumers (LCS). The relation strength between terms, calculated by CORDER, was then used to refine these extended queries. Finally, the Hyperspace Analogue to Language (HAL) model and Information Flow (IF) method were applied to expand these refined queries further.

Initial experimental results on a corpus of 500 books from Amazon demonstrated that this approach successfully identified relevant books for users, even when given vague queries. In several cases, it outperformed both Google and Amazon’s own book search in delivering accurate results.

Team

Dawei Song

Marc Eisenstadt

Chris Denham

Jianhan Zhu