A Secure and Communication-Efficient Federated Recommendation Method With Blockchain

Due to the significant advantages of federated learning (FL) in privacy protection, federated recommendation systems (FedRSs) have garnered increasing attention by enhancing recommendation performance through local data training. However, most current FedRSs adopt a client-server communication architecture, which may lead to communication overload and single points of failure. Additionally, clients may face challenges due to limited communication resources and malicious attacks. To address the above challenges, we propose a Blockchain-assisted Federated learning method for Recommendation, called BFedRec, suitable for recommendation systems with high communication efficiency requirements. Specifically, BFedRec achieves the aggregation and distribution of recommendation models through a blockchain system, reducing reliance on central servers and alleviating communication bottlenecks and single points of failure. On this basis, BFedRec applies an innovative FL method that trains recommendation models directly on low-rank parameters to achieve efficient and secure data aggregation and distribution. Moreover, the flexibility of this aggregation and distribution strategy allows for scalable applications in other fields, such as blockchain-enabled software-defined network (SDN) management in on-chain and off-chain communication networks. Experimental results demonstrate that BFedRec outperforms existing methods on real datasets, significantly improving communication efficiency while effectively enhancing system security and robustness.