Knowledge discovery in relational database: extraction of n-ary relationships

Abstract

Presents an automatic process for extracting the n-ary relationships from an operating relational database. This database reverse engineering process performs (i) the extraction of n-ary relationship tables, and (ii) the deduction of cardinalities. The first step is based on querying a view of the data dictionary and some working tables. The second step is based on the automatic generation of a set of SQL queries for each n-ary relationship table. We focus on the realistic assumptions that there are no constraints on the uniqueness of the attribute names, and dependencies are not supposed to be known a priori. A PRO*C program has been implemented; it enables us to automatically build SQL queries via dynamic SQL. We also study the influence of the number of key attributes and the number of records upon the performance of our process.