Coalition Formation Among Rational Information Agents

Abstract

Information agents can be an important tool for information-gathering and query- answering for the expanding WWW service as well as the large number of existing autonomous databases. Such agents behave like active intelligent front-ends of the stand-alone information systems. Information agents may work as individuals trying to satisfy thier own query-answering, i.e., a set of given information search tasks. However, they must cooperate effi ciently with one another in order to gather information in non-local domains. In this paper we present an approach for cooperation and coalition formation among information agents for heterogeneous databases. In order to deal with the required association autonomy of these agents, we have developed a special decentralized coalition formation mechanism. It allows individually rational cooperation among the agents for information search. The semi-automatic creation of a local terminological information model enables each agent to hide local schema data as well as allowing for automated search-task processing. Knowledge-based productions from information search-tasks are sets of terminological interdatabase dependencies. These productions are obtained by classifying the search term of the task according to the local information model. An underlying cooperation convention in this federative agent system is that only members of a fi xed coalition are mutually committed to providing the availability of all of the local information that they have used for building this particular coalition.