Jing Du

PhD Candidate

University of New South Wales

I am a PhD candidate at School of Computer Science and Engineering, the University of New South Wales, supervised by Prof. Lina Yao and Prof. Wenjie Zhang. Before that, I received my M.S. and B.S. degree from the School of Computer Science, Northwestern Polytechnical University, in 2020 and 2017, respectively. My research interest is information retrieval and data mining, especially their applications in recommender systems. I have published 3 first-author papers in referred journals and conference proceedings, including ACM SIGIR, IEEE Internet of Things, etc. I served as an external reviewer of ACM SIGKDD, PAKDD, ACM SIGIR, and the reviewer of Knowledge-based system (KNOSYS) and PPNA. I received several awards and honors, including SIGIR Student Travel Grant, Outstanding Master Thesis of Northwestern Polytechnical University, Zhong Hang Talent Scholarships, etc. 

With the proliferation of multi-source data, the types of services that can be provided are increasing. To solve the problem of information overload, people put forward the recommendation system to help people find the items to meet their needs easily. Generally, the recommendation system trains the user profile based on its historical behavior information, and then makes personalized recommendation for each user.  Recommender system has become a basic application in the field, which provides reliable choices for users. However, the existing works have some drawbacks that have not been overcome yet. First, poor generalization to scenarios where tasks are loosely correlated. Second, an item or a user, which has never been identified before, needs to be recognized. Third, when learning a new task, the parameters of networks are updated, and the knowledge of the old task will be covered. To tackle the problems above, we consider three key objectives, listed as follows: 1) Multi-task Recommender System. Our aim is to bridge these gaps by presenting a neural multi-task learning model to explicitly distinguishes task-specific features from task-shared features to reduce the impact caused by weak correlation between tasks. 2) Self-supervised Recommender System. We propose socially-aware self-supervised learning for cold-start recommendation, where cold users can be modeled in the same way as warm users. 3) Meta Recommender System. We study and develop the long/short-term interest of users and migrated the knowledge of related users to guide the prediction and recommendation in the target object.