HeteroGraphRec: A heterogeneous graph-based neural networks for social recommendations

Recommender systems in social networks are widely used for connecting users to their desired items from a vast catalog of available items. Learning the user's preferences from all the possible sources of information in an extensive, multi-dimensional social network is one of the main challenges when building such recommenders. Graph Neural Networks have been gaining momentum in recent years and have been successful when dealing with large-scale graphs, and they can be applied to social networks with some modifications. In this research, we propose the HeteroGraphRec, which provides social recommendations by modeling the social network as a heterogeneous graph and utilizing GNNs with attention mechanisms to intelligently aggregate information from all sources when building the connections between user to user, item to item, and user to item. The HeteroGraphRec can gather information about the user's connections (friendships, trust network), item interaction history, and item similarities to attain rich information about the preferences. To evaluate the HeteroGraphRec, we use three real-world benchmark datasets and demonstrate that the proposed HeteroGraphRec achieves superior performance compared to ten other state-of-the-art social recommender systems. We extensively analyze the HeteroGraphRec model to illustrate the effectiveness by changing the embedding dimensions of the users and items. We also show the interpretability of our model by examining each component of the model's contribution. The results show that the HeteroGraphRec is robust and can consistently perform better than the baseline systems. (C) 2021 Elsevier B.V. All rights reserved.

Automated summary

HeteroGraphRec is a social recommender system that models the social network as a heterogeneous graph and intelligently aggregates information using GNNs with attention mechanisms. Research shows that HeteroGraphRec outperforms top social recommender systems, demonstrating strong robustness and performance superiority.