Towards Effective Recommendation: Auxiliary Data and Personalized Hyper-parameter Learning
Chen Ma
McGill University
Montreal, QC, Canada
Via https://zoom.us/j/96417215332?pwd=SEMxdUt5VVdnY2hRdWdlTi80TmdHQT09
Abstract
With the development of Internet services and mobile devices, Internet users can easily access a large number of online products and services. Although this growth provides users with more available choices, it is also difficult for users to pick up one of the most favorite items out of plenty of candidates. To reduce information overload and satisfy the diverse needs of users, personalized recommender systems come into being and play more and more important roles in modern society. In this talk, two directions will be presented to better understand the user preference and improve the recommendation performance. Firstly, I will present how to effectively utilize auxiliary data, such as temporal orders of items. In particular, dedicated modules will be introduced to model the user interest in a fine-grained manner. Secondly, I will present how to build effective models to make good use of user-item interactions. Specifically, an interesting solution--adaptive/personalized hyper-parameter learning mechanism will be presented. Lastly, I will summarize this talk with a broader vision of recommendation techniques beyond recommendation accuracy.
Biographical Sketch
Chen Ma is a Ph.D. candidate in the School of Computer Science at McGill University, supervised by Prof. Xue (Steve) Liu. His research is on the intersection of recommender systems and deep learning. The central theme driving his research is searching for effective auxiliary information, powerful models, adaptive/personalized hyper-parameters, and fairness-aware recommendation results. His work has been deployed in a real-world Mobile App Store with millions of monthly active users.