Effective optimization in expert systems

In this research, we create new techniques for the following three areas of expert systems: information acquisition, optimization of choices, and explanations. It is desirable for an expert system to intelligently acquire new information and avoid asking irrelevant questions. We show how an expert system can learn the relevancy of questions. The expert system analyzes its reasoning after each usage of the system to determine relevancy of questions. This learned insight is used to minimize the number of questions asked of the user in future sessions. We also show how a futile line of questioning can be avoided. Expert systems generally determine feasible solutions to problems. Such solutions may not be optimal with respect to implementation cost. We show how expert systems can be developed that determine an optimal solution to a problem. Our technique involves the decomposition of large problems into smaller subproblems. An expert system needs the ability to explain its decision making. We show how to build an explanation facility that uses a bottom-up approach. Simple concepts are explained first and then used to explain more difficult ones. We have integrated the three techniques in an expert system that solves the important and difficult problem of industrial chemical exposure management. The system is called OCHEM (Optimal Chemical Hazard Exposure Management). OCHEM is a multilevel expert system whose input criteria include government regulations, company policy, economic factors, and time constraints. The OCHEM system intelligently questions a user and determines an optimal solution for the user's hazardous chemical exposure problem. The recommendations of the OCHEM system are then explained to the user using simple English sentences.