Rectangular decomposition of signature file: Design issues and performance evaluation

Abstract

Signature file acts as a filter to eliminate the unqualified data items for an information request. The contribution of this paper is two-fold. First, we propose a methodology for structured database. The main idea of decomposition amounts to dividing the relations of database into subrelations called ’rectangle’. Owing to the reduction of storage overhead, rectangular decomposition has a direct benefit in a reduction of false drop probability. Second, we extend the use of rectangular decomposition to partitionning signature file called Super-Rectangular Organisation SRO. The choice of the criteria to cluster signatures is based upon the relationship between the associated objects. Our main concern is to study the performance of this method in terms of this ability to reduce the search space and false drop probability. We also perform simulation experiments by comparing several methods like S-tree and Quick filter. The results show that the SRO improves performance significantly in term of retrieval time and false drop probability.