The paper proposes a federated heterogeneous graph neural network (FedHGNN) for privacy-preserving recommendations. The authors suggest that the Heterogeneous Information Network (HIN) is partitioned into private HINs stored on the client side and shared HINs on the server. The proposed model can collaboratively train a recommendation model on distributed HINs without compromising user privacy. The model outperforms existing methods by up to 34% in HR@10 and 42% in NDCG@10 under an acceptable privacy budget.
Publication date: 19 Oct 2023
Project Page: https://doi.org/XXXXXXX.XXXXXXX
Paper: https://arxiv.org/pdf/2310.11730
