Discovering concept clusters by decomposing databases

摘要

This paper introduces an approach of discovering concept clusters by decomposing databases. This approach is the fundamental one for developing DBI which is one of sub-systems of the GLS discovery system implemented by us. A key feature of this approach is the formation of concept clusters or sub-databases through analysis and deletion t of noisy data in decomposing a database. Its development is based on the concept of Simon and Ando's near-complete decomposability that has been most explicitly used in economic theory. In this approach, the process of discovering concept clusters from databases is a process based on incipient hypothesis generation and refinement, are many kinds of learning methods, in which the methods of data-driven and knowledge-driven are included, are cooperatively used in multiple learning phases, so that a more robust, general discovery system can be developed.