Predicting Users' Purchasing Behaviors Using Their Browsing History

Tieke He,Hongzhi Yin,Zhenyu Chen,Xiaofang Zhou,Bin Luo

DOI: https://doi.org/10.1007/978-3-319-19548-3_11

2015-01-01

Abstract:Some E-commerce giants (e.g., Amazon and Jingdong) with abundant purchasing data achieve highly accurate recommendations, since people have to pay for their choices and their purchasing behaviors are more qualified and valid for capturing users' needs and preferences than other types of users' behavior data (e.g., browsing). However, there is not enough users' purchasing data available for most of small and medium-size E-commerce sites as well as some newly established E-commerce sites. In this paper, we aim to alleviate the sparsity of users' purchasing data by exploiting users' browsing data which is more sufficient. The low validity and reliability of users' browsing data raises great challenge for accurately predicting users' purchasing behaviors since there are many factors leading to users' browsing behaviors. To this end, we propose a novel semi-supervised method to make the most of both high-quality purchasing data and low-quality browsing data to predict users' purchasing behaviors. Specifically, we first use a small amount of purchasing data to supervise the model training of browsing data, and then integrate the results into the item-based collaborative filtering method. We conduct extensive experiments on a real dataset, and the experimental results show the superiority of our method by achieving 25% improvements over traditional collaborative-filtering methods.