Published in Proceedings of the Twentieth Innovative Applications of Artificial Intelligence Conference, September 1, 2007. 6 pages. © 2008 Association for the Advancement of Artificial Intelligence. Conference proceedings also available online at http://www.aaai.org/Press/Proceedings/proceedings.php.
Legacy system data models can interoperate only if their syntactic and semantic differences are resolved. To address this problem, we have developed the Intelligent Mapping Toolkit (IMT), which enables mixed-initiative mapping of meta-data and instances between relational data models. IMT employs a distributed multi-agent architecture so that, unlike many other efforts, it can perform mapping tasks that involve thousands of schema elements. This architecture includes a novel federation of matching agents that leverage case-based reasoning methods. As part of our pre-deployment evaluation for USTRANSCOM and other DoD agencies, we evaluated IMT’s mapping performance and scalability. We show that combinations of its matching agents are more effective than any that operate independently, and that they scale to realistic problems (i.e., that involve thousands of schema elements).