A PROPOSED ARCHITECTURE FOR SELF-ADAPTIVE EXPERT SYSTEMS

摘要

The pre-built knowledge of traditional expert systems is only capable of limited responses to changes in the operating environment. If the data input is imperfect, a traditional system may fail to reach any rational conclusions. In this paper, we introduce the concept of self-adaptability to the inference process of an expert system, and propose a model that is capable of handling unexpected user input effectively and efficiently. Such a system can formulate operational knowledge on the move for inference. With this self-adaptive capability, an expert system can reach useful conclusions, even when the input data is insufficient. The architecture of the proposed system encodes domain knowledge with semantic networks. It also defines four types of adaptation, namely, condition knowledge adaptation, operational knowledge adaptation, conclusion knowledge adaptation, and presentation adaptation, and focuses on how the first three contribute to the adaptive capability of the system. In addition, to enable a self-adaptive expert system to effectively produce better conclusions, two entropy-based measuring mechanisms are proposed: one minimizes the information loss during knowledge adaptation, while the other selects the best attribute relation during the generation of operational knowledge. We have proved that a self-adaptive expert system based on this architecture can always reach a regular conclusion or an abstract conclusion, which is a more meaningful conclusion by automatically modifying its operational knowledge in response to user feedback during the inference process, even in unexpected situations.