Fusion Graph Convolutional Collaborative Filtering

Recently, Graph Neural Network (GNN) has been proved to be an efficient technique to solve the problem of graph-structured data, and many graph-based methods of recommendation have shown noticeably good performances. However, many approaches use pure GNN layers as the encoder of the nodes, which we think may limit the performance of the model. In this paper, we propose Fusion Graph Convolutional Collaborative Filtering (FGC-CF) which uses DeepWalk and graph convolutional layers to be the encoder of nodes to enhance the capability of the node encoder. For better modeling the similarity of user and item, we involve the local inference of the ESIM [3] to obtain the user representations by considering the interacted items, and the item representations by considering the interacted users. We conduct experiments on four datasets and the results not only show the remarkable performance of FGC-CF but also prove the necessity of using DeepWalk and ESIM local inference technologies.