Recently, the paper results of The 38th Annual AAAI Conference on Artificial Intelligence (AAAI 2024) were announced. The paper“LGMRec: Local and Global Graph Learning for Multimodal Recommendation”guided by Prof.GuohuiLiand authored by his PhD student ZhiqiangGuo was accepted.

Multimodal recommendation has become a fundamental application on today’s main stream multimedia platforms. The paper analyzes two major issues in existing multimodal recommendation tasks: (1) Shared updating of user ID embeddings leads to the coupling of collaborative and multimodal signals; (2) Lack of exploration of robust global user interests to mitigate the interaction sparsity faced in local interest modeling.

In response to these issues, the paper proposes a multimodal recommender (LGMRec) guided by local and global graph learning to jointly model users’local and global interests. The method first uses a local graph embedding module to independently learn the collaborative and modal representations of users and items with local to pological relationships. Then, a global hypergraph embedding module is further designed to capture the global representations of users and items by modeling global dependencies. By combining the global representations obtained in the hypergraph embedding space with the two decoupled local representations, the accuracy and robustness of multimodal recommendations are improved. Experiments conducted on three public benchmark datasets show that the LGMRec model outperforms various advanced recommendation baselines, demonstrating its effectiveness in modeling local and global user interests.

AAAI is one of the oldest and most comprehensive top international academic conferences in the field of artificial intelligence (CCF A-class conference), with a total of 12,100 papers submitted to this conference, 9,862 effective submissions, and 2,342 papers finally accepted, with an acceptance rate of about 23.75%. Prof.GuohuiLiis the corresponding author of this academic paper.