CMBSR: Contrastive Multi-Behavior Social Recommendation via Graph Neural Networks

Most of the traditional recommendation algorithms are based on single-behavior interaction, ignoring the multiple-interaction behaviors displayed by users in reality. In the real world, there is far more than one type of interaction between users and items. For example, in e-commerce platforms, users may perform various operations such as collecting, adding a shopping cart, and purchasing, which reflect users' different intentions and preferences for items. At the same time, multi-behavior recommendations still seriously suffer from the problem of limited supervision signals. This paper proposes a Contrastive Multi-Behavior Social Recommendation model named CMBSR to solve the above problems. Based on multi-behavior social rec-ommendation algorithm, we add three comparison mechanisms: multi-behavior comparison, user social interaction comparison and item knowledge comparison, which make full use of multi-behavior information, social information and item knowledge to capture user complex preferences and behavior patterns. Experimental results show that CMBSR performs significantly better than other advanced algorithms on multiple datasets. Especially on the Tmall dataset, CMBSR achieves a 4.958% and 5.923% performance improvement on HR@10 and NDCG@10 compared with the sub-optimal result, respectively.