Federated Social Recommendation with Graph Neural Network

Recommender systems have become prosperous nowadays, designed to predict users’ potential interests in items by learning embeddings. Recent developments of the Graph Neural Networks (GNNs) also provide recommender (RSs) with powerful backbones learn embeddings from a user-item graph. However, only leveraging interactions suffers cold-start issue due difficulty data collection. Hence, current endeavors propose fusing social information alleviate it, which is recommendation problem. Existing work employs GNNs aggregate both links and simultaneously. they all require centralized storage item users, leads privacy concerns. Additionally, according strict protection under General Data Protection Regulation, may not be feasible future, urging decentralized framework recommendation. As result, we design federated system for task, rather challenging because its heterogeneity, personalization, requirements. To this end, devise novel Fe drated So cial G raph neural network ( FeSoG ). Firstly, adopts relational attention aggregation handle heterogeneity. Secondly, infers user using local retain personalization. Last but least, proposed model pseudo-labeling techniques sampling protect enhance training. Extensive experiments on three real-world datasets justify effectiveness completing protection. We are first proposing best our knowledge.