A personalized cross-domain recommendation with federated meta learning

Cross-domain Recommendation (CDR) is an effective method for solving the cold-start problem in traditional recommendation systems. This technique transfers information from the source domain to the target domain, enabling personalized recommendations for cold-start users. Despite the prevalence of existing CDR methods, they often rely on a common preference bridge to convey information to all users, which may not capture complex and diverse preferences unique to each user. Additionally, these methods require direct interaction of user information between clients and servers, which can pose personal privacy concerns. To address these issues, we propose a personalized cross-domain preference bridge for each user using federated meta learning. Our approach trains a common preference bridge for all users using locally stored data and model information with federated learning. Then, a meta-learning network generates a personalized preference bridge for each user based on the common bridge. This approach improves the effectiveness of CDR and ensures user privacy. We conducted extensive experiments on two large datasets to demonstrate the efficiency of our proposed method. The results show that our model outperforms other models in addressing the cold-start problem in recommendation systems.