Martin Atzmueller, Joachim Baumeister, Frank Puppe
When knowledge systems are deployed into a real-world application, then the maintenance and the refinement of the knowledge are essential tasks. Many existing automatic knowledge refinement methods only provide limited control and clarification capabilities during the refinement process. Furthermore, often assumptions about the correctness of the knowledge base and the cases are made. However, such assumptions do not necessarily hold for real-world applications. In this paper, we present a novel interactive approach for the refinement of knowledge bases: Subgroup mining is used to discover local patterns that describe factors potentially causing incorrect behavior of the knowledge system. The approach is supplemented by introspective subgroup analysis techniques in order to help the user with the interpretation of the refinement recommendations proposed by the system.
Subjects: 10. Knowledge Acquisition; 12. Machine Learning and Discovery
Submitted: Feb 13, 2006