知識獲得のための知識表現「専門家モデル」

抄録

This paper describes Expert Model, a knowledge representation for knowledge acquisition and Pre-post method, a knowledge acquisition method based on the expert model. The expert model consists of operations which represent small tasks in human experts' problem solving processes. We have prepared generic operations for effective model building. We assumed that there is some representation technique in real knowledge bases because knowledge engineers build them, although the production rules are in a very simple and general form. We have obtained seven types of generic operations by analyzing real knowledge bases which are written for diagnosis problems in production rules. The pre-post method has two interview strategies. It stimulates a human expert to remember by ask what operations are done before and after an operation. It extracts knowledge about operations efficiently, according to the operation type. This paper also introduces a knowledge acquisition support system, EPSILON/One, based on the expert model and the pre-post method. It has been implemented on PSI (Personal Sequential Inference machine) in ESP (Extended Self-contained Prolog). It consists of a human interface for interview, a knowledge elicitation module, a knowledge refinement module, and a knowledge translation and inference module. The interface supports graphic multi-windows and a mouse. The knowledge elicitation module extracts knowledge in the form of the expert model by the pre-post method. The knowledge refinement module supports to detect a lack of knowledge. The knowledge translation and inference module translates the expert model into knowledge in the form of ESP language and supports to evaluate the knowledge in inference processes.