Modeling the generative mechanism of personalized preferences from latent groups: A hierarchical nonparametric Bayesian method

Abstract

Personalization is among the most prominent marketing strategies for facing online and offline retailers. The basis for personalization is learning the customer’s personalized preferences, but the sparsity of customer behavioral data makes it difficult to calculate personalized preferences and predict behaviors. Assuming that customer preferences are influenced by the latent group, we propose a Hierarchical Nonparametric Bayesian method (HNB) to model the generative mechanism of personalized preferences from latent groups. The proposed HNB employs a hierarchical and nonparametric structure based on the Dirichlet process, and constructs a combined model to analyze personalized preferences and latent groups. In the context of online shopping, HNB not only mitigates data sparsity with information from the latent groups, but also identifies the meaning of preference in customer purchase history and discovers latent groups accurately. Our experiments also show the number of preferences and latent groups is calculated automatically from the data, understand the generative mechanism of personalized preference from latent groups, and predict personalized purchases based on preferences.