FedCIA: Federated Collaborative Information Aggregation for Privacy-Preserving Recommendation

TL;DR

FedCIA tackles privacy concerns in recommender systems by shifting federation from parameter aggregation to collaborative information aggregation, exchanging item similarity matrices instead of model parameters. Grounded in graph signal processing, it shows that the average of local item similarity filters approximates the global collaborative information, enabling lossless or reduced-information aggregation and preserving personalized embeddings. The framework supports both parameter-based and parameter-free models, with a practical training pipeline that includes local DP noise on uploaded similarities and a mean-squared-error refinement step using the global similarity. Empirical results on five real-world datasets demonstrate that FedCIA consistently outperforms state-of-the-art federated rec systems in accuracy, convergence, personalization, and scalability, while maintaining privacy protections.

Abstract

Recommendation algorithms rely on user historical interactions to deliver personalized suggestions, which raises significant privacy concerns. Federated recommendation algorithms tackle this issue by combining local model training with server-side model aggregation, where most existing algorithms use a uniform weighted summation to aggregate item embeddings from different client models. This approach has three major limitations: 1) information loss during aggregation, 2) failure to retain personalized local features, and 3) incompatibility with parameter-free recommendation algorithms. To address these limitations, we first review the development of recommendation algorithms and recognize that their core function is to share collaborative information, specifically the global relationship between users and items. With this understanding, we propose a novel aggregation paradigm named collaborative information aggregation, which focuses on sharing collaborative information rather than item parameters. Based on this new paradigm, we introduce the federated collaborative information aggregation (FedCIA) method for privacy-preserving recommendation. This method requires each client to upload item similarity matrices for aggregation, which allows clients to align their local models without constraining embeddings to a unified vector space. As a result, it mitigates information loss caused by direct summation, preserves the personalized embedding distributions of individual clients, and supports the aggregation of parameter-free models. Theoretical analysis and experimental results on real-world datasets demonstrate the superior performance of FedCIA compared with the state-of-the-art federated recommendation algorithms. Code is available at https://github.com/Mingzhe-Han/FedCIA.

Figures (5)

• Figure 1: Illustration of two aggregation paradigms in federated recommendation algorithms.
• Figure 2: The illustration of the FedCIA framework. We only illustrate a single client for simplicity. The red part indicates private information and the yellow part indicates global information.
• Figure 3: The learning curve for different aggregations on MF in Ml-100k and Ml-1m datasets.
• Figure 4: Distribution of item embeddings during different stages of federated learning in different aggregations.
• Figure 5: The comparison for different aggregations on heterogeneous scenarios in Ml-100k dataset.

Theorems & Definitions (1)

• Theorem 1