A Decomposition Based Induction Model for Discovering Concept Clusters from Databases

Abstract

This paper presents a decomposition based induction model for discovering concept clusters from databases. This model is a fundamental one for developing DBI which is one of sub-systems of the GLS discovery system implemented by us. A key feature of this model is the formation of concept clusters or sub-databases through analysis and deletion of noisy data in decomposing databases. Its development is based on the concept of Simon and Ando's nearcomplete decomposability that has been most explicitly used in economic theory. In this model, the process of discovering concept clusters from databases is a process based on incipient hypothesis generation and refinement, and many kinds of learning methods are cooperatively used in multiple learning phases for performing multi-aspect intelligent data analysis as well as multi-level conceptual abstraction and learning, so that a more robust, general discovery system can be developed.