On July 10, the 20th issue of the "Tsinghua Software Forum" series of academic activities invited Philip S.Yu (Yu Shilan), distinguished professor of the University of Illinois at Chicago, to give an academic report entitled "On Recommendations via Large Multi-modal Models". School of Software president Wang Jianmin, Associate Professor Wen Lijie, Song Shaoxu and more than 20 teachers and students attended the event. Nearly 100 people, including teachers and students inside and outside the university and research fields, participated in the conference through Tencent Conference and video account of School of Software of Tsinghua University. The academic forum was chaired by Wang Jianmin.

Professor Philip S.Yu gave a report

Professor Philip S.Yu introduced the challenges in the construction and application of recommendation systems, and shared the cutting-edge work on modeling and learning from data from different sources using large multimodal models. The first task is to use graph collaborative filtering to recommend items to users. This work builds dichotomies using the browsing and interaction records of users and project sets, and observes that the associations between users and items in project sets can be used to construct auxiliary edges. The trained model takes into account the above user-item relationships. Through the training graph encoder and the weight matrix, the items recommended by the recommendation system to the user will take into account different types of association relationships. The second work focuses on temporal association prediction in social networks. This work starts from the dynamic evolution of association relationships in social networks, and adds the prediction dimension of start and end time to the relationship prediction problem. This method first uses temporal network to train and predict the changes of correlation in social graph, and then uses time-dependent topic model to predict correlation, which can be well applied to practical problems such as dynamic recommendation in social graph.

In the question session, the teachers and students asked questions around the recommendation system, data mining, temporal data and other interest points, as well as the application of large models and other hot issues, Professor Philip S.Yu analyzed and answered in detail one by one.

On-site of the report

【BIOS of Philip S.Yu】

Philip S.Yu is a distinguished professor at the University of Illinois at Chicago. His main research areas include big data mining, data fusion and anonymous data, especially in the model construction and application of recommendation system for different data sources.