Multi-Layer Collaborative Graph with BPR Similarity Embedding for Recommender System

Learning high-level representations for users and items is the main objective of modern recommender systems (RS). Since data in RS inherently has graph structure the graph neural networks (GNN) have been widely used for feature extraction of RS. To take advantage of GNN in representation learning we present a new efficient method, named as Multi-Layer Collaborative Graph with BPR Similarity Embedding (MLCGBPRSE) by applying multi-layer graph convolutional network for obtaining high-order neighborhood information in propagation process. Moreover, for better information exploiting we present an auxiliary loss function to capture user-user and item-item similarities as well. The experimental results indicate the superiority of our proposed model in both recall and NDCG compared to the state-of-the-art approaches.