Using concept hierarchies to enhance user queries in web-based information retrieval

The effectiveness of Internet search engines is often hampered by the ambiguity of user queries and the reluctance or inability of users to build less ambiguous multi-word queries. Our system, ARCH, is a client-side Web agent, which incorporates domain-specific concept hierarchies together with interactive query formulation in order to automatically produce a richer and therefore less ambiguous query. Unlike traditional relevance feedback methods, ARCH assists users in query modification prior to the search task. ARCH uses the domain knowledge inherent in Web-based classification hierarchies such as Yahoo, combined with a user's profile information, to add just those terms likely to improve the match with the user's intent. The goal of the system is therefore to meet the user's information needs by closing the gap between the user's stated query and the actual intent of the search. We present a detailed evaluation of the query enhancement in ARCH, comparing enhanced and non-enhanced queries over a range of topics. Our results show that concept-based query enhancement in ARCH leads to significantly higher precision for ambiguous queries without sacrificing recall.