A Knowledge Acquisition System for Price Change Rules

Abstract

Knowledge acquisition is the process of extracting and organizing knowledge from one source and storing it in some other location such as a knowledge base. Our research developed a new approach to knowledge acquisition concerning motor fuel pricing and implemented it in the Knowledge Acquisition System for Price ChangE Rules (KASPER) software system. Store managers want to understand the pricing strategies at competing stores or brands. The main goal of our research is to provide decision rules with high predictive accuracy on unseen data that may explain why a store or brand made a price change in a speci c category. These decision rules should relate prices at one store to those at other stores or brands in the same city. Our approach is able to generate directional and categorical price change rules. The approach can use brand-based or distance-based store-to-store relations or use brand- to-brand relations. KASPER was applied to data from four cities to generate decision rules from these relations. We tested the decision rules on unseen data and found that most decision rules had high predictive accuracy in cases where the price changes tend to

uctuate more. Our approach was more e ective in the two cities where price changes of varied sizes occur than in the two cities where price changes are of consistent, small sizes. We found that high variability of price changes allows the system to match corresponding behaviours more e ectively.