An Attention-Based Approach to Client-Selected Federated Recommendations

With the rapid development of data analytics and cloud computing, how to enhance the efficiency of data processing while protecting user privacy has become a key challenge. This study proposes a novel privacy-enhanced lossless federated corecommendation framework (AttFedRec++), which integrates the attention mechanism with homomorphic encryption. By adaptively evaluating the dynamics of the task and the current state of the client, AttFedRec++ is able to optimize the resource allocation in federated learning, thus improving the computational efficiency. Meanwhile, this framework adopts Paillier-DJN homomorphic encryption scheme to ensure the security of data during communication. Experimental results on the standard dataset MovieLens show that AttFedRec++ can effectively accelerate the convergence speed of the model without sacrificing the recommendation accuracy.