An automatic technique for detecting type conflicts in database schemes

Abstract

Type conflict detection, consisting in singling out different database scheme representations denoting the same portion of the reality, is a key point in the development of cooperative information systems. Indeed, when constructing global views of available data, conflicting objects coming from distinct input schemes are collapsed into a unique object. When the amount of metadata involved in the integration process is huge, traditional manual approaches to type conflict detection are difficult to apply. This paper proposes an automatic, probabilistic approach to the detection of type conflicts in database schemes. The method we are proposing here is based on considering pairs of objects belonging to different schemes and measuring their similarity. To this end, object structures as well as object neighborhoods are analyzed to verify similitudes and differences. The method is encoded in a simple fixpoint computation defined on a graph structure we associate with each database scheme. The method is shown to run in polynomial time in the size of involved schemes. A number of examples show the suitability of our techniques to effectively detect type conflicts.