Latent Topic Analysis for Predicting Group Purchasing Behavior on the Social Web.

Group-deal websites, where customers purchase products or services in groups, are an interesting phenomenon on the Web. Each purchase is kicked off by a group initiator, and other customers can join in. Customers form communities with people with similar interests and preferences (as in a social network), and this drives bulk purchasing (similar to online stores, but in larger quantities per order, thus customers get a better deal). In this work, we aim to better understand what factors influence customers' purchasing behavior for such social group-deal websites. We propose two probabilistic graphical models, i.e., a product-centric inference model (PCIM) and a group-initiator-centric inference model (GICIM), based on Latent Dirichlet Allocation (LDA). Instead of merely using customers' own purchase history to predict purchasing decisions, these two models include other social factors. Using a lift curve analysis, we show that by including social factors in the inference models, PCIM achieves 35% of the target customers within 5% of the total number of customers while GICIM is able to reach 85% of the target customers. Both PCIM and GICIM outperform random guessing and models that do not take social factors into account.