ENTROPY-BASED MEASURES FOR PARTITIONING THE DOMAIN OF AN INTERPRETATION IN DESCRIPTION LOGICS

Abstract

A basic feature of description logic-based information systems is binary relations between objects. Concept learning in these systems need to take advantages of those relationships for granulating partitions of the domain of an interpretation (i.e., a description logic-based information system). In this paper, we study a method for partitioning the domain of an interpretation in description logics using bisimulation. Apart from information gain measure, we propose a new measure, called first depth information gain, to choose selectors for dividing blocks in partitions. Our results show that this measure is valuable and it is an important foundation for bisimulation-based concept learning methods in description logics.