A Formal Framework for Integrating Inconsistent Answers from Multiple Information Sources

In any environment of multiple information sources it is practically unavoidable that information sources would overlap; that is, describe the same portion of the real world. And when information sources overlap, it is practically unavoidable that information would be inconsistent; that is, describe diierently the same portion of the real world. In this report we deene a formal framework for resolving inconsistencies that are detected in the process of processing queries against a collection of information sources. This framework establishes basic database concepts, such as schemes, instances, views, queries and integrity constraints, as well as new concepts, such as view equivalence, scheme mappings, multi-databases, multi-database queries and inconsistency. Altogether, it provides the formal foundations necessary for addressing issues of information integration and inconsistency resolution. Our approach to the integration of inconsistent answers is based on information goodness, a measure for estimating the proximity of stored information to the ideal information. We propose to maintain for each information source a goodness basis: a set of basic views whose goodness has been established and are expected to be useful for inferring the goodness of anticipated queries. When a query is answered inconsistently by individual information sources we propose to (1) infer the goodness of each individual answer from the appropriate goodness basis, and (2) integrate the individual answers in an answer of the highest goodness. We sketch the architecture of a software agent, called Harmony, that will implement our approach.