Towards a Trust-Aware Item Recommendation System on a Graph Autoencoder with Attention Mechanism

Recommender Systems provide users with recommendations for potential items of interest in applications like e-commerce and social media. User information such as past item ratings and personal data can be considered as inputs of these systems. In this study, we aim to utilize a trust-graph-based Neural Network in the recommendation process. The proposed method tries to increase the performance of graph-based RSs by considering the inferred level of trust and its evolution. These recommendations will not only be based on the user information itself but will be fueled by information about associates in the network. To improve the system performance, we develop an attention mechanism to infer a level of trust for each connection in the network. As users are likely to be influenced more by those whom they trust the most, our method might lead to more personalized recommendations, which is likely to increase the user experience and satisfaction.