On meta levels of an organized society of KDD agents

Abstract

Modeling of KDD (Knowledge Discovery in Databases) process constitutes an important and new research area of KDD, that is, meta levels of the KDD process, including formal specification of the process, its planning, scheduling, controlling, management, evolution, and reuse. The key issue is how to increase both autonomy and versatility of a KDD system. Our methodology is to create an organized society of KDD agents. This means (1) to develop many kinds of KDD agents for different discovery tasks; (2) to use the KDD agents in multiple learning phases in a distributed cooperative mode; (3) to manage the society of KDD agents by multiple meta-control levels. Based on this methodology, a multi-strategy and cooperative KDD system, which can be imagined as a softbot and is named GLS (Global Learning Scheme), has being developing by us. This paper focuses on the meta control levels for increasing both autonomy and versatility of the KDD system.