Speaker Profile

KOU Gang, Member of CPPCC National Committee, Deputy Director of Xiangjiang Laboratory, Dean of Research Institute of Big Data at Southwestern University of Finance and Economics, and Vice President of Systems Engineering Society of China. As a national distinguished professor, he is also a recipient of National Science Fund for Distinguished Young Scholars and receives special government allowances from State Council.

Professor Kou has led numerous major research projects, including key projects from National Natural Science Foundation of China (NSFC). He has published over 200 papers in prestigious outlets, including Science, Nature portfolio journals, and UTD24-listed publications (such as ISR and JOC), as well as at top-tier conferences like ICML, AAAI, and KDD. His work has garnered over 20,000 citations, with an H-index of 77. His research has earned him numerous awards, including Ministry of Education's First Prize for Excellence in Scientific Research. More than ten of his policy proposals have been adopted or recognized at the highest levels of government, receiving commendations from national leaders.

Abstract

Recommendation systems rely on rich historical interaction data to deliver accurate results. In practice, however, these systems are often hampered by the "cold start" problem, where a lack of interaction data for new users or items leads to severe data scarcity. To address these challenges, we first propose a multi-view graph contrastive learning method that integrates both attribute and structural information. Its adaptive contrastive learning module dynamically adjusts the mutual information between views, allowing it to extract valuable signals from both attribute and structural data.

To further combat data scarcity, we introduce a multi-view fusion framework that leverages this method to integrate user social networks and item semantic correlations, effectively addressing the challenge of sparse data. For cold start scenarios, we design a bidirectional online distillation mechanism. This mechanism enables knowledge transfer between a "content-enhanced collaborative embedding network" and a "content-based embedding network," achieving an adaptive fusion of content information and collaborative signals. This approach effectively resolves the cold start problem and significantly enhances recommendation performance.

Date/Time: September 15, 15:30 p.m.

Venue: Room 0204, 0 Teaching Building, Jiuliu Campus