SHGCN: Socially Enhanced Heterogeneous Graph Convolutional Network for Multi-Behavior Prediction

In recent years, multi-behavior information has been utilized to address data sparsity and cold-start issues. The general models capture multiple behaviors of users make the representation relevant features more fine-grained informative. However, most current recommendation methods neglect exploration social relations between users. Actually, users’ potential connections are critical assist them in filtering multifarious messages, which may be one key for tap deeper into interests. Additionally, existing usually focus on positive (e.g. click , follow purchase ) tend ignore value negative unfollow badpost ). this work, we present a Multi-Behavior Graph (MBG) construction method based user relationships, then introduce novel socially enhanced behavior-aware graph neural network behavior prediction. Specifically, propose Socially Enhanced Heterogeneous Convolutional Network (SHGCN) model, utilizes heterogeneous convolution module effectively incorporate achieve precise addition, aggregation pooling mechanism is suggested integrate outputs different layers, dynamic adaptive loss (DAL) presented explore weight each behavior. experimental results datasets e-commerce platforms (i.e., Epinions Ciao) indicate promising performance SHGCN. Compared with powerful baseline, SHGCN achieves 3.3% 1.4% uplift terms AUC Ciao datasets. Further experiments, including model efficiency analysis, DAL ablation confirm validity enhancement.