Estimating the Quality of Databases

With more and more electronic information sources becoming widely available, the issue of the quality of these often-competing sources has become germane. We propose a standard for specifying the quality of databases, which is based on the dual concepts of data soundness and data completeness. The relational model of data is extended by associating a quality specification with each relation instance, and by extending its algebra to calculate the quality specifications of derived relation instances. This provides a method for calculating the quality of answers to arbitrary queries from the overall quality specification of the database. We show practical methods for estimating the initial quality specifications of given databases, and we report on experiments that test the validity of our methods. Finally, we describe how quality estimations are being applied in the Multiplex multidatabase system to resolve cross-database inconsistencies.