Discovery of first-order regularities in a relational database using ofine candidate determination

Abstract

In this paper, we present an algorithm for the discovery of first order clauses holding in an relational database in the framework of the nonmonotonic ILP setting [1]. The algorithm adopts the principle of offline candidate determination algorithm used for mining association rules in large transaction databases [4]. Analoguous to the measures used in mining association rules, we define a support and a confidence measure as acceptance criteria for discovered hypothesis clauses. The algorithm has been implemented in C with an interface to the relational database management system INGRES. We present and discuss the results of an experiment in the KRK domain and conclude.