Mixture models and their applications in market basket analysis

This thesis studies the latent phenomenon in customers’ shopping history by leveraging advanced techniques in Statistics, Machine Learning, and Optimization. The term ‘latent phenomenon’ refers to a phenomenon that is informative but not directly observable. Quantitatively inferring latent phenomena, referred to as the latent variable modeling, is of vital importance to various disciplines, including Statistics, Psychology, Business, and so forth.

In our first study, we define latent phenomena as customers’ shopping motivations. We propose latent variable models to estimate the motivations and the motivation change behind customers’ shopping trips. These motivations are represented by customer’ purchases, and we show that estimating these motivations can assist retailers in customer segmentation and product recommendation.

In our second study, we define latent phenomena as customer profiles. We propose latent variable models to identify the customer profiles that are associated with higher (or lower) monetary value, as well as the customer profiles that are more (or less) sensitive to coupons. The estimated customer profiles help retailers in terms of customer management and determining the target audience for marketing campaigns.