Towards Personalized Privacy: User-Governed Data Contribution for Federated Recommendation

Federated recommender systems (FedRecs) have gained significant attention fortheir potential to protect user's privacy by keeping user privacy data locallyand only communicating model parameters/gradients to the server. Nevertheless,the currently existing architecture of FedRecs assumes that all users have thesame 0-privacy budget, i.e., they do not upload any data to the server, thusoverlooking those users who are less concerned about privacy and are willing toupload data to get a better recommendation service. To bridge this gap, thispaper explores a user-governed data contribution federated recommendationarchitecture where users are free to take control of whether they share dataand the proportion of data they share to the server. To this end, this paperpresents a cloud-device collaborative graph neural network federatedrecommendation model, named CDCGNNFed. It trains user-centric ego graphslocally, and high-order graphs based on user-shared data in the server in acollaborative manner via contrastive learning. Furthermore, a graph mendingstrategy is utilized to predict missing links in the graph on the server, thusleveraging the capabilities of graph neural networks over high-order graphs.Extensive experiments were conducted on two public datasets, and the resultsdemonstrate the effectiveness of the proposed method.