PRIVACY PRESERVING RECOMMENDATION SYSTEM USING FEDERATED LIGHTGCN WITH TEMPORAL MODELING

Abstract

Recommendation systems play an important role in modern day digital applications such as eCommerce, video streaming, and social networks by allowing users to find relevant information amidst large datasets. Classical recommendation methods such as Collaborative Filtering and Matrix Factorization suffer from certain limitations such as sparse data, cold start problems, scalability issues, and privacy concerns due to centralization of datasets. To overcome these problems, we propose an ap proach to building a privacy-preserving recommendation system based on Federated Light Graph Convolutional Networks along with temporal modeling and federated weighted aggregation. The proposed approach uses recent interactions for more sig nificant influence on user preferences and uses weighted federation for aggregating models based on the size of client datasets. The experiments were conducted on the MovieLens 32M dataset, where performance is measured using Precision@10, Re call@10, and NDCG@10. From our experimental evaluations, the proposed method has shown improvements of 13%, 16%, and 12% in Precision@10, Recall@10, and NDCG@10, respectively, compared to a baseline federated model. At the same time, the proposed method preserves privacy of users’ personal data. It is found that the integration of LightGCN, temporal modeling, and federated learning offers an effective and scalable solution to modern recommendation systems.