| When searching the web, a user strives to find useful documents. Web search engines have been shown to have relatively low coverage, as well as other problems, that limit their ability to return useful documents. One solution is use of a metasearch engine: a tool that sends user queries to multiple search engines and combines the results to increase coverage. Unfortunately, a metasearch engine also has problems, mostly due to the lack of direct control of the underlying search engines, that limit its ability to locate and identify useful results.; To improve the ability of users to find useful results when searching the web, we present the architecture of a preference-based metasearch engine Inquirus 2, which utilizes explicit user preferences in the form of a category, such as “personal homepages” or “research papers”. Inquirus 2 demonstrates five architectural improvements that enhance the ability to locate and identify useful documents, and to increase performance. The architectural improvements are: an incremental user interface, need-based source selection, source and category-specific query modification, selective downloading of results, and need-based scoring.; A user study was performed to measure the effectiveness of the architectural components and to improve our understanding of user judgments of usefulness as related to topical relevance and document category. This user study demonstrated a bounding relation between the levels of user judgments of topical relevance and usefulness, as well as confirming the effectiveness of our architecture.; We also present the Query Modification Learning Procedure (QMLP), a procedure that automatically learns category-specific and source-specific query modifications, along with experimental results confirming the effectiveness of the procedure. |
