On using historical update information for instance identificationin federated databases

摘要

To support database interoperability in federated databasessystems, it is critical to be able to identify (potentially) equivalentdata instances from individual autonomous database components. Since thecomponents in a federation are autonomous, their data may be updatedasynchronously, viz., modifications to a real world entity may becaptured in different databases at different times; the authors termthis effect update heterogeneity. Existing approaches largely base datainstance similarity identification only on current attribute/propertyvalues; in the face of update heterogeneity, this is inadequate. Theypresent an approach to address the problem of update heterogeneity inthe federated databases context. They employ a probabilistic model,which utilizes historical database update information to estimate thedegree of similarity between candidate data instances from differentdatabase components. They employ transaction history (log) informationto this end, which is typically already available in the componentdatabase systems. They have experimentally implemented and tested thisapproach within the context of a prototype experimental federateddatabases system, FeXpress