This project will (1) complete the implementation of the existing design for an experimental expert retrieval assistance system in the workstation environment; (2) test the effectiveness of the system through experimentation and analysis; and (3) depending on the results of the analysis, devise more refined and comprehensive search models and assistance techniques. The techniques used incorporate such functionality as (1) estimating absolute and relative search relevance based on several spectra of matching criteria; (2) automatic ranking of documents by relevance using existing tools from current Boolean-based systems; and (3) optimization of search strategies based on user relevance feedback. These new techniques will be integrated with a collection of assistance aides in the current system including (1) automatic execution of search strategies in different databases including translation from a common command language; (2) search strategies derived automatically from user's natural language phrases; and (3) recall estimation based on both formal search features and empirical search results. These models and techniques, termed "smart Boolean", will advance information retrieval from an intuitive art toward a rational quantified decision-making process.
