A Restriction-Based Approach to Generalizations

Abstract

Generalizations, also known as contrast patterns, are in the core of many learning systems. A key component to automatically find generalizations is the predicate to select the most important ones. These predicates are usually formed by restrictions that every generalization must fulfill. Previous studies are mainly focused on the types of generalizations, each one associated to a particular predicate. In this paper, we shift the focus from predicates to restrictions. Restrictions are analyzed based on a set of intuitions that they materialize. Additionally, an analysis of the restrictions used in a large collection of existing generalizations suggests interesting conclusions.