Combining View Integration and Schema Clustering to Improve Database Design
The aim of this paper is to provide a coherent combination of two well-known techniques improving database design : view integration and schema clustering. View integration methods suggested in the literature tend to address mainly the following problems: terminology problems, class definition overlappings, constraint contradictions, different representations of concepts. However two other problems remain unsolved. The first one is concerned with view comparison, which may lead to a very costly process requiring, for two views containing about n objects, up to n2 elementary comparisons. This paper proposes a clustering technique for solving this problem. The clustering is based on an automatic process, reducing the number of elementary comparisons to n2/4 in the worst case and n in the best case. The second problem is related to view integration validation. The view integration process leads to a global schema amalgamating all the initial views. View integration validation consists mainly in checking that this objective is met. This paper proposes an application of automatic clustering to view integration permitting to recover the initial views. In order to validate view integration, the clustering algorithm requires the application of several semantic distances taking into account different integration situations. We propose to perform the clustering process using successively several distances until a satisfying partitioning is obtained. The two techniques have been applied to several case studies leading to empirical but very promising results.
COMYN-WATTIAU, I., AKOKA, J. et KEDAD, Z. (1998). Combining View Integration and Schema Clustering to Improve Database Design. Dans: BDA’98. Quatorzièmes Journées Bases de Données Avancées. Laboratoire PriSM, Université de Versailles, pp. 59-80.