Using metarules to integrate knowledge in knowledge based systems. An application in the woodworking industry.

摘要

The current study addresses the integration of knowledge obtained from Data Mining structures and models into existing Knowledge Based solutions. It presents a technique adapted from commonKADS and spiral methodology to develop an initial knowledge solution using a traditional approach for requirement analysis, knowledge acquisition, and implementation. After an initial prototype is created and verified, the solution is enhanced incorporating new knowledge obtained from Online Analytical Processing, specifically from Data Mining models and structures using meta rules. Every meta rule is also verified prior to being included in the selection and translation of rules into the Expert System notation. Once an initial iteration was completed, responses from test cases were compared using an agreement index and kappa index.;The problem domain was restricted to remake and rework operations in a cabinet making company. For Data Mining models, 8,674 cases of Price of Non Conformance (PONC) were used for a period of time of 3 months.;Initial results indicated that the technique presented sufficient formalism to be used in the development of new systems, using Trillium scale. The use of 50 additional cases randomly selected from different departments indicated that responses from the original system and the solution that incorporated new knowledge from Data Mining differed significantly. Further inspection of responses indicated that the new solution with additional 68 rules was able to answer, although with an incorrect alternative in 28 additional cases that the initial solution was not able to provide a conclusion.