Personalized Federated Recommendation for Cold-Start Users via Adaptive Knowledge Fusion

Yichen Li, Yijing Shan, YI LIU, Haozhao Wang, wangshi.ww, Yi Wang, Ruixuan Li

Track: User modeling, personalization and recommendation

Keywords: Federated Learning, Recommendation System, Cold-Start User

Abstract:

Federated Recommendation System (FRS) usually offers recommendation services for users while keeping their data locally to ensure privacy. Currently, most FRS literature assumes that fixed users participate in federated training with personal IoT devices (e.g., mobile phones and PC). However, users may come incrementally, and it is unfeasible to retrain the whole FRS with the new participating user due to the expensive training overheads and the negligible global knowledge gain brought by a small number of new users. To guarantee the quality service for these new users, we take a dive into the federated recommendation for cold-start users, a novel scenario where the new participating users can directly achieve a promising recommendation without overall training with all participating users by leveraging both transferred knowledge from the converged warm clients and the knowledge learned from the local data.

Nevertheless, how to efficiently transfer knowledge from warm clients remains controversial. On the one hand, cold clients may introduce new sparse items, causing a distribution shift from the item embedding converged on warm clients. On the other hand, the user information from warm clients is required to match cold users for a collaborative recommendation, but directly sharing user information is a violation of privacy and unacceptable. To tackle these challenges, we propose an efficient and privacy-enhanced federated recommendation for cold-start users (FR-CSU) that each client can adaptively transfer both user and item knowledge from warm clients separately and implement recommendations with local and transferred knowledge fusion. Specifically, each cold client will train a mapping function locally to transfer the aligned item embedding. Meanwhile, warm clients will maintain a user prototype network in a FedAvg manner that provides privacy-friendly yet effective user information for cold users. Finally, a linear function system will fuse the transferred and local knowledge to improve the recommendation. Extensive experiments show that FR-CSU achieves superior performance compared to state-of-the-art methods.

Submission Number: 1616