2022 Volume 73 Issue 2 Pages 54-69
A marketing policy called the ”Membership Stage System” is widely used in retail business. A membership stage provides benefits to customers such as shopping points when a customer's annual cumulative purchase amount exceeds a certain threshold and the customer's stage is raised a level. As a result, the company is not only able to promote the customer's willingness to purchase, but it can also obtain the purchasing history data, thereby enabling high-quality customer analysis. The most fundamental analysis is to infer the difference of purchasing characteristics between member stages and to construct different clustering models for each member stage. However, when the clustering models are learned independently for each membership stage, it is not possible to compare the obtained clusters between membership stages. In this study, we propose a new analytical method and its learning algorithm to analyze differences in cluster distribution between membership stages. Through demonstrating the proposed model applied to an actual data set of purchasing history data on a membership stage system, the effectiveness of our proposal is clarified.
