A symbolic approach using feature construction capable of acquiring information/knowledge for building expert systems

摘要

This paper illustrates a technique useful for developing a knowledge-base of information to use in an expert system. The proposed approach employs a popular machine-learning algorithm along with a method for forming a finite number of features or conjuncts of at most n primitive attributes. We illustrate our procedure by examining the qualitative information represented in a set of consumer survey responses typically used to create measures of consumer confidence. After proposing a premise related to certain assumptions implicit in current models used to develop a quantitative measure of consumer confidence, our results show the certain assumptions regarding the construction of a consumer confidence measure appear to be appropriate for the time period we examined. We also demonstrate how to use our approach to form certain demographic descriptions of consumers from the survey data. Researchers may use this procedure to develop a knowledge-base of information that is capable of being easily understood by humans.