A Framework for Information Integration with Uncertainty

Abstract

Uncertainty management and information integration have been challenging issues in AI and database research. The literature is vast and rich on either of these two issues, however, they have not been studied simultaneously in the same setting. In this work, we make a first attempt and propose a framework for information integration with uncertainty, which uses the information source tracking(IST) model [9] as the underlying certainty model. The IST model is an extension of the relational data model in which every tuple t is annotated with (a set of) fixed length vectors, called agent vectors, representing the (human or sensor) agents which confirmed t or contributed to it. Our framework consists of a dynamic collection of autonomous but cooperating IST databases, called the information sources or sites, in which each relation r is annotated with a site vector, indicating which sites contributed to the definition of r. We extend the relational algebra from the basic IST model accordingly to manipulate agent and site vectors. We also extend the reliability calculation algorithm from the basic model to compute the certainty of each answer tuple as a function of the reliabilities of the contributing agents and sites. We have developed a running prototype of the proposed framework for which we mainly used SQL programming for query rewriting and manipulation of agent and site vectors.