Dynamic classificational ontologies for discovery in cooperative federated databases

Abstract

A Cooperative Federated Database System (CFDBS) is an information sharing environment in which units of information to be shared may be substantially structured, and participants are actively involved in sharing activities. We focus on the problem of shared ontology for the purpose of discovery in the CFDBS context. We introduce the concept and mechanism of the Dynamic Classificational Ontology (DCO), which is a mediator to help participants identify and resolve ontological similarities and differences. A DCO contains top level knowledge about information units exported by information providers, along with classificational knowledge. By contrast with fixed hierarchical classifications, the DCO builds domain specific, dynamically changing classification schemes; it specifically contains knowledge about overlap among information units. Information providers contribute to the DCO when information units are exported, and the current knowledge in the DCO is in turn utilized to guide export and discovery of information. At the cost of information providers' cooperative efforts, this approach supports much more systematic discovery than that provided by keyword based search, with substantially greater precision and recall. An experimental prototype of the DCO has been developed, and applied and tested to improve the precision and recall of Medline document searches for biomedical information sharing.