Cooperative Answering through Controlled Query Relaxation

AS KNOWLEDGE STORAGE SYStems grow in size and complexity, it becomes unreasonable to expect that users always know enough about the data and how the data are organized to construct accurate and useful queries. Even when a system has syntactic utilities for building queries, users continue to compose queries that contain misconceptions, are too general, or are bound to fail. Cooperative-answering strategies seek to detect such situations and collaborate automatically with users to find the information that the users are seeking. They provide users with additional information,1,2 intermediate answers,3 qualified answers,4 or alternative queries.5 One form of cooperative behavior involves providing associated information that is relevant to a query.6–9 Relaxation—generalizing a query to capture neighboring information—is a means to obtain possibly relevant information. A cooperative-answering system described here uses relaxation to identify automatically new queries that are related to the user’s original query. For large databases, many relaxations might be possible. Therefore, the system incorporates methods to control relaxation: user constraints restrict the relaxation, and heuristics order the relaxed queries. The methods extend to incorporate semantic information from electronic thesaurii. The system has been implemented as a prototype query constructor. These methods combine to form a declarative, domain-independent method that lets users tailor queries through the exploration of semantic relationships. The methods work in concert with the user to provide a queryanswering system that answers not just the user’s literal query, but the intended query.