Federated Neural Collaborative Filtering. (arXiv:2106.04405v1 [cs.IR])

In this work, we present a federated version of the state-of-the-art Neural
Collaborative Filtering (NCF) approach for item recommendations. The system,
named FedNCF, allows learning without requiring users to expose or transmit
their raw data. Experimental validation shows that FedNCF achieves comparable
recommendation quality to the original NCF system. Although federated learning
(FL) enables learning without raw data transmission, recent attacks showed that
FL alone does not eliminate privacy concerns. To overcome this challenge, we
integrate a privacy-preserving enhancement with a secure aggregation scheme
that satisfies the security requirements against an honest-but-curious (HBC)
entity, without affecting the quality of the original model. Finally, we
discuss the peculiarities observed in the application of FL in a collaborative
filtering (CF) task as well as we evaluate the privacy-preserving mechanism in
terms of computational cost.