Predicting Time-bounded Purchases During a Mega Shopping Festival

Abstract

Shopping festivals such as Black Friday are an important period for online retail businesses. Shopping behaviors during these periods tend to be rushed and time-bounded due to many short-lived promotions, which makes it challenging for businesses to expect traffic and provision their service. In this paper, we examine the click action logs of five million shoppers from a large online retail market during a mega sale event in China and present a model of shopper behavior. Our model takes into account a wide range of features including the product category, time of day, and click sequences. Our model could identify precursors of purchase from as early as two weeks prior to the shopping festival. These findings help online retailers better prepare for future events by efficiently classifying shoppers and thereby reducing the operational costs associated with provisioning any rushed shopping moments.