B-FedGNN: Blockchain-Coordinated, Privacy-Preserving Federated Graph Neural Networks for Social-Commerce Recommendations

Social-commerce recommendations leverage user interactions and social relationships to enhance shopping experiences, yet existing centralized approaches raise significant privacy concerns and scalability limitations. We propose a novel privacy-preserving framework that combines Federated Graph Neural Networks (FedGNN) with blockchain technology to enable scalable social-commerce recommendations across retail networks. Our approach tackles three critical challenges: (1) preserving user privacy while utilizing social-graph structures; (2) enabling cross-retailer collaboration without exposing competitive data; and (3) achieving real-time performance at scale.We introduce a hierarchical federated-learning architecture in which Graph Neural Networks are trained locally on user devices and retail nodes, capturing social-influence patterns without exposing raw data. A purpose-built blockchain coordinates the learning process through smart contracts, ensuring transparent model aggregation and immutable audit trails. To address scalability, we implement novel graph-partitioning algorithms, asynchronous model-aggregation protocols, and an optimized consensus mechanism supporting up to 10000 transactionss<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−1</sup>.Our system employs differential-privacy techniques and secure multi-party computation to protect both individual user data and graph topology. Experiments on three large-scale data sets show that our approach retains 94.3% of centralized accuracy while processing more than one million recommendations per minute across 1000 retail nodes. Compared with existing solutions, we cut communication overhead by 73% and improve recommendation latency by 5×, while providing formal privacy guarantees (ε =0.1, δ =10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−5</sup>).The framework enables retailers to collaboratively improve recommendation quality without compromising user privacy or leaking proprietary data, paving the way for truly decentralized social-commerce ecosystems.