A multi-agent architecture for cooperative query answering

Many people have considered the idea of query relaxation as a means of supplying an approximate answer when an exact match to a database query could not be found. By supplying an answer that nearly matches the original request, it is hoped that the user may still be able to find a satisfactory match. Much of the work to date, however, has not examined the different "relaxation needs" of different users. This paper proposes a solution by using a series of user profiling agents to extract information about the user from their patterns of searching behaviour. Information is extracted by the various agents in a mostly invisible manner, by observing the behaviour which users undertake as part of their normal searching practice. Multiple agents work in a dynamic manner, obtaining and combining their information as needed by other agents or by the system as a whole. The agents make intelligent inferences about the user and continuously seek and respond to feedback, modifying and refining their assumptions.