Brand purchase prediction based on time-evolving userbehaviors in e-commerce

摘要

Purchase prediction is a key function in the e-commerce recommendation system. Existingworksusually focus on item-level purchase prediction, which faces two issues of high cost and low accuracy.In this paper, we study brand purchase prediction by exploring behaviors, which may lead tobrand purchases.We make three progresses. (1)We analyze a real world e-commerce data frommultiple angles. Focusing on users' brand purchases, we find behaviors' evolution with time andbehaviors' interaction. (2)For differentbehaviors,we extract different time-evolving features thatcan serve as indicators of users' brand purchase. (3) We use a logistic regression-based modelby adjusting the parameters of time-evolving feature and others in two different scenarios (thepromotion purchase prediction and the daily purchase prediction) to construct two experiments.The experiment results show that themodel using three types features performs the best in bothscenarios, and the time-evolving feature plays the most important role among them. (4)We distinguishthe feature importance in different scenarios. Based on the importance, we find thatusers' purchases in the promotion scenario are likely to be impulsive,while purchases in the dailyscenario aremore likely to be influenced by users' activities.