Abstract

Current computer networks o er the possibility of accessing and sharing widely distributed information. The SIMS architecture 2] o ers a system for sharing information between multiple, heterogeneous, and distributed information agents. A SIMS information agent is given a model of some domain, models of the contents of information sources (IS) that refer to such domain, and a description of how IS concepts relate to those of the domain model. Then, it uses these models to nd the most appropriate set of information sources that satisfy a given query. These models are expressed in a rich knowledge representation system (Loom 3], a description logic). This paper addresses the problem of automatically reconciling the model of an agent with that of its information sources (other agents, data or knowledge bases, etc), and adapt these models to the evolution of the IS's. The main idea is to analyze the actual extensions (ie, the current instances) of the IS's, checking the consistency of this data with the de nitions in the domain and IS models. This processing will generate novel concepts that will prove useful in improving the accuracy of these models and the e ciency of the system. A prototype for model reconciliation has been implemented using a SIMS mediator that accesses relational databases. The outline of this extended abstract is as follows. First, we present the di erent types of new concepts that may appear when integrating IS's. Then, we describe how to generate (learn) useful descriptions of these concepts. Third, we show how such concepts can be used to re ne the models, adapt to changes in the contents of the IS's and improve the e …

Date

September 21, 1994

Authors

Jose Luis Ambite, Craig A Knoblock

Journal

Proc. Workshop on Intelligent Information Agents