Entity Identi cation in Database Integration: An Evidential Reasoning Approach

Entity identiication is the problem of matching object instances from diierent databases which correspond to the same real-world entity. In this paper, we present a 2-step entity identiication process in which attributes for matching tuples may be missing in certain tuples, and thus need to be derived prior to the matching. To match tuples, we require identity rules which specify the conditions to be satissed by a pair of tuples, from diierent databases, before they can be considered as modeling the same real-world entity. We also introduce ILFD's (instance-level functional dependencies) as a form of inference rules which derive the missing identifying attributes. In order to provide more interesting integrated results to the users, we allow both identity rules and ILFD's to contain indee-niteness represented as necessary and possible support information. Based on support logic programmingg2], we develop an approach to perform reasoning on the local databases using identity rules and ILFD's.