FEDGNNREC: A FEDERATED GRAPH NEURAL NETWORK FRAMEWORK FOR PRIVACY-PRESERVING RECOMMENDATIONS
Recommender systems are common in today's online platforms to give personalised recommendations RS depending on the interaction of the users. Nevertheless, the conventional centralised recommendation methods demand gathering user information in central servers, which poses a serious privacy risk and a higher risk of information leakage. Concurrently, most of the traditional models do not reflect the structural relations that exist between the user and item interactions. In order to resolve these problems, this paper will suggest a FGNN framework, in which privacy-preserving recommendation is provided. The method proposed is a federated learning, which is merged with GNN to allow decentralised training whilst maintaining the relational structure of data in terms of interaction. All the clients build a local user-item graph and learn GNN-based embedding without distributing raw data. The aggregating server receives updates on the model by a federated averaging mechanism in order to construct a global model. The experimental results with Amazon review data indicate that the proposed federated GNN has a correlation of 0.9176, RMSE of 0.4021, and MAE of 0.2135, which is better than multiple centralised and base methods. Full privacy is also guaranteed by the model since the user data is not transferred to any other clients. These findings demonstrate the usefulness of federated graph-based learning in order to have secure and accurate recommendation systems.