演讲人简介:Dr. Tong Wang is an Assistant Professor in the Department of Business Analytics at the University of Iowa. She received her Ph.D. in Computer Science from the Massachusetts Institute of Technology in 2016. Dr. Wang has been working on developing novel machine learning methods and applying them to business problems and social good problems. Her work has been published in both top machine learning conferences & journals and business journals. Her research on crime data mining is the second place winner in "Doing Good with Good OR” at INFORMS 2015, and has been implemented by the New York Police Department. Her crime pattern detection algorithm has been reported in multiple media including wired.com and Wikipedia. Dr. Wang's team developed a solution to a FICO XML challenge in 2018, which won the FICO recognition award. Her work on interpretable machine learning have won several best paper awards at data mining related conferences and workshops under INFORMS.
报告摘要:In recent years, e-commerce platforms have expanded into an important new product domain of financial products that includes payment, credit, investment, and insurance products. However, due to the scarcity of data in this new product domain, online platforms face challenges in predicting users' purchase behavior. In this paper, we study whether we can ``transfer'' knowledge learned from the existing consumer goods domain to benefit the prediction in the domain of the financial products. With data provided by one of the largest online shopping platforms in China, we develop machine learning solutions to enable knowledge transfer. We show that users' prior browsing and shopping history in consumer goods can significantly improve the prediction accuracy of users' purchases of mutual funds for both the existing-user and the new-user scenarios. In addition, we study the heterogeneous prediction performance lifts on users with different socioeconomic statuses and investment risk preferences. Results show that information from the consumer goods domain has a higher prediction performance lift on users in the high socioeconomic group. Finally, we compare the effect of different sources of information on predicting users' purchases of mutual funds. We find that users' browsing and shopping history for consumer goods are more predictive than their profile features. Our findings and methods will be valuable to both the financial industry and online platforms that seek to expand their product domains.