FDRS: Federated Diversified Recommender System Based on Heterogeneous Graph Convolutional Network

Currently, researchers are working to improve the accuracy of recommender systems. However, recommender systems ignore users' diversity requirements in recommendation lists, thus reducing user satisfaction. Additionally, in the big data environment, users' data information is scattered in different enterprises or organizations, but these enterprises or organizations are reluctant to share users' data information due to the existing laws and regulations, which leads to the recommendation process facing the problem of the sparsity of users' available data. Based on the above research, we propose a novel framework of a federated diversified recommender system, i.e., FDRS, which is a diversified recommendation method following the principle of vertical federation learning. Specifically, FD RS first constructs a heterogeneous users-items graph, then uses a heterogeneous graph convolutional network to learn user representations based on heterogeneous information, and finally, based on vertical federation learning, FDRS performs model parameter aggre-gation, rating prediction, and diversity recommendation. We compare our approach with existing recommendation methods on the Epinions dataset. The experimental results show that our proposal greatly alleviates the accuracy and diversity dilemmas while preserving data privacy.