A Multiagent Referral System for Expertise Location

Referral systems are designed to simplify person-to-person communication by helping find relevant humans with whom to communicate regarding a specified topic. Such systems maintain and explore an informal person-to-person social network, which provides knowledge about the relevance of potential correspondents to a user’s information needs. Like our approach, current referral systems are based on software agents. However, current systems rely exclusively on collections of email messages and newsgroup postings. Messages and postings only partially indicate the sender’s expertise and interests. As a result, previous approaches yield poor referrals for requests outside of the domain of the given collection. By contrast, we develop a system that (1) combines techniques from information retrieval and adaptive user modeling to refine the social network according to the user’s needs, and (2) allows agents to unintrusively exchange explicit profile information to add coverage regarding a user’s expertise. Moreover, our system is a truer multiagent system than previous approaches in that it explicitly relies on exchanges among the agents to build and refine the social network. A prototype of our system has been implemented and results of its use are presented.