Rule mining using many-sorted logic

Abstract

Inductive Logic Programming (ILP) is used in relational data mining to discover rules in first order logic, given data in multiple relations. This form of data mining has to be distinguished from market basket analysis where the data comes from a single relational table. Although ILP addresses the problem of dealing with data from multiple relational tables, the fact remains that the efficiency of inferring rules in first order logic is significantly less than that of many-sorted logic. Further, many sorted logic is a closer reflection of the real world of objects that belong to sorts, in the presence of a sort hierarchy. We propose a new approach to ILP using many-sorted logic that is more computationally efficient than the approach based on unsorted first order logic.