Transfer Learning Based on Probabilistic Latent Semantic Analysis for Analyzing Purchase Behavior Considering Customers' Membership Stages

In recent years, it has become common to analyze purchase history data and take advantage of the effect on business policies. In this study, the authors focus on the case of a company that is introducing a membership stage system. Probabilistic latent semantic analysis (PLSA) is well-known as an analytical model for analysis of co-occurrence of variables in data. However, the relationship between customers and items for the customer purchase behavior of each stage based on PLSA has not shown good performance. The purchase behaviors may be slightly different between customer stages, and accordingly, the purchase behavior of customers in different stages should be represented by a similar but different model. In addition, the higher the membership stage, the fewer customers there are; therefore, it becomes difficult to accurately understand the features of the customers' purchase behavior within high membership stages. In this study, the authors propose a learning algorithm that utilizes the estimated parameters of the behavior model based on the PLSA model at a lower-stage, to estimate the parameters of models at a higher-stage to which few people belong. Moreover, numerical simulation experiments are conducted and compared to actual purchase history data to confirm the performance of the proposed method.