E-commerce purchase prediction approach by user behavior data

Abstract

while e-commerce has grown quickly in recent years, more and more people are used to utilize this popular channel to purchase products and services on the Internet. Therefore, it becomes very important for shopping sites to predict precisely which items their customers would buy so as to increase sales or improve customer satisfaction. Traditional algorithms such as Collaborative Filtering, has been very popular in predicting users' preferences in movie, book, or music recommendation areas, but they face the problem that rating data is very sparse or even not available in shopping domain. Compared to the small amount of ratings in e-commerce shopping sites, the quantity of user clicking data is abundant and also contains sufficient information about users' purchase preferences. Therefore, in this paper we propose a prediction method based on probability statistics making use of user clicking behavior data. To evaluate the proposed approach, we use the data set provided by Ali Mobile Recommendation Competition held in 2015 which consisted of the huge amount of the clicking behavior log from 10,000 mobile users in one month. The experimental results show that our proposed method significantly alleviates the problem of data sparseness which traditional algorithms fail to deal with.