A method of acquiring strategies and classifying knowledge from trace data

It is important for a knowledge processing system to have a meta-level inference mechanism for reducing problem search space. The authors have developed such a mechanism for KAUS (Knowledge Acquisition and Utilization System) and applied it to control system design, chemical structure design, etc. However, it is difficult for the domain experts to give meta-level knowledge. This is partly because the object knowledge is not well structured in their minds. This paper proposes a method to classify knowledge and to acquire strategies for improving system performance from trace data. Results of two experiments are given to show the effectiveness of the proposed method.