CIF-HGR: A privacy-preserving and collaborative framework for cross-institutional federated heterogeneous graph recommendation in energy IoT

Abstract

In the era of big data and the Energy Internet of Things (eIoT), recommendation systems are crucial for mitigating information overload and enhancing user experience. However, fragmented and heterogeneous data across different platforms and devices hinders accurate and diverse recommendations. To address this limitation, we propose CIF-HGR, a collaborative and privacy-preserving federated heterogeneous graph recommendation framework tailored for eIoT. CIF-HGR integrates three major components: (1) heterogeneous graph construction and adaptive learning to extract high-quality representations from large-scale eIoT data; (2) a consortium blockchain-based collaboration layer offering privacy-preserving identity authentication and secure session management; and (3) a federated heterogeneous graph recommendation scheme that employs differential privacy and contribution-based incentives to ensure fairness, efficiency, and sustainability.

Extensive experiments confirm that our method outperforms recent baselines in both single- and multi-institution eIoT environments. In the single-institution scenario on Amazon, CIF-HGR improves NDCG@10 from 0.2487 (LightGCN) to 0.2639 and Coverage@10 from 0.1034 to 0.1191. Under federated aggregation, our optimized scheme raises Coverage@10 from 0.1647 (FedKD) to 0.1682 and remains competitive with FedMR’s NDCG@10 (0.2940 vs. 0.2936). Moreover, CIF-HGR maintains an attack success rate below 0.85 at $ϵ=1.0$, underscoring its efficacy in balancing accuracy, coverage, and privacy.