Merging Relational Views: A Minimization Approach

Abstract

Schema integration is the procedure to integrate several inter-related schemas to produce a unified schema, called the mediated schema. There are two major flavors of schema integration: data integration and view integration. The former deals with integrating multiple data sources to create a mediated query interface, while the latter aims at constructing a base schema, capable of supporting the source schemas as views. Our work builds upon previous approaches that address relational view integration using logical mapping constraints. Given a set of data dependencies over the source schemas as input, our approach produces a minimal information-preserving mediated schema with constraints, and it generates output mappings defining the source schemas as views. We extend previous approaches in several aspects. First, schema minimization is performed within a scope of Project-Join views that are information preserving and produce a smaller mediated schema than in existing work. Second, the input schema mapping language is expressive enough for not only query containment but also query equivalence. Third, source integrity constraints can be seamlessly incorporated into reasoning. Last but not least, we have evaluated our implementation over both real world data sets and a schema mapping benchmark.