A Hybrid Approach to Natural Language Web Search

A Hybrid Approach to Natural Language Web Search Jennifer Chu-Carroll, John Prager, Yael Ravin and Christian Cesar IBM T.J. Watson Research Center P.O. Box 704 Yorktown Heights, NY 10598, U.S.A. {jencc,jprager,ravin,cesar}@us.ibm.com Abstract We describe a hybrid approach to improv- ing search performance by providing a natural language front end to a traditional keyword-based search engine. The key component of the system is iterative query formulation and retrieval, in which one or more queries are automatically formulated from the user’s question, issued to the search engine, and the results accumulated to form the hit list. New queries are gener- ated by relaxing previously-issued queries using transformation rules, applied in an order obtained by reinforcement learning. This statistical component is augmented by a knowledge-driven hub-page identi- fier that retrieves a hub-page for the most salient noun phrase in the question, if possible. Evaluation on an unseen test set over the www.ibm.com public web- site with 1.3 million webpages shows that both components make substantial contri- bution to improving search performance, achieving a combined 137% relative im- provement in the number of questions cor- rectly answered, compared to a baseline of keyword queries consisting of two noun phrases. 1 Introduction Keyword-based search engines have been one of the most highly utilized internet tools in recent years. Nevertheless, search performance remains unsatis- factory at most e-commerce sites (Hagen et al., 2000). Librarians and search professionals have tra- ditionally favored Boolean keyword search systems, which, when successful, return a small set of rele- vant hits. However, the success of these systems crit- ically depends on the choice of the right keywords and the appropriate Boolean operators. As the pop- ulation of search engine users has grown beyond a small dedicated search professional community and as these new users are less familiar with the contents they are searching, it has become harder for them to formulate successful keyword queries. To improve search performance, one can improve search engine accuracy with respect to fixed keyword queries, or provide the search engine with better queries, those more likely to retrieve good results. While there is much on-going work in the IR community on the former topic, we have taken the latter approach by providing a natural language search interface and automatically generating keyword queries that uti- lize advanced search features typically unused by end users. We believe that natural language ques- tions are easier for users to construct than keyword queries, thus shifting the burden of optimal query formulation from the user to the system. Such ques- tions also eliminate much of the ambiguity of key- word queries that often leads to poor results. Fur- thermore, the methodology we describe may be ap- plied to different search engines with only minor modification. To transform natural language input into a search query, the system must identify information perti- nent for search and utilize it to formulate keyword queries likely to retrieve relevant answers. We de- scribe and evaluate a hybrid system, RISQUE, that adopts an iterative approach to query …

Content truncated to 500 words