Abstract:Collecting and training over sensitive personal data raise severe privacy concerns in personalized recommendation systems, and federated learning can potentially alleviate the problem by training models over decentralized user <a class="link-external link-http" href="http://data.However" rel="external noopener nofollow">this http URL</a>, a theoretically private solution in both the training and serving stages of federated recommendation is essential but still <a class="link-external link-http" href="http://lacking.Furthermore" rel="external noopener nofollow">this http URL</a>, naively applying differential privacy (DP) to the two stages in federated recommendation would fail to achieve a satisfactory trade-off between privacy and utility due to the high-dimensional characteristics of model gradients and hidden <a class="link-external link-http" href="http://representations.In" rel="external noopener nofollow">this http URL</a> this work, we propose a federated news recommendation method for achieving a better utility in model training and online serving under a DP <a class="link-external link-http" href="http://guarantee.We" rel="external noopener nofollow">this http URL</a> first clarify the DP definition over behavior data for each round in the life-circle of federated recommendation <a class="link-external link-http" href="http://systems.Next" rel="external noopener nofollow">this http URL</a>, we propose a privacy-preserving online serving mechanism under this definition based on the idea of decomposing user embeddings with public basic vectors and perturbing the lower-dimensional combination coefficients. We apply a random behavior padding mechanism to reduce the required noise intensity for better utility. Besides, we design a federated recommendation model training method, which can generate effective and public basic vectors for serving while providing DP for training participants. We avoid the dimension-dependent noise for large models via label permutation and differentially private attention modules. Experiments on real-world news recommendation datasets validate that our method achieves superior utility under a DP guarantee in both training and serving of federated news recommendations.

DPLCF: Differentially Private Local Collaborative Filtering

Differentially Private User-based Collaborative Filtering Recommendation Based on K-means Clustering

Privacy Preserving Point-of-Interest Recommendation Using Decentralized Matrix Factorization.

Enforcing Differential Privacy for Shared Collaborative Filtering

An Algorithm for Efficient Privacy-Preserving Item-Based Collaborative Filtering

Fairness-aware Differentially Private Collaborative Filtering

SDRM-LDP: A Recommendation Model Based on Local Differential Privacy

An Effective Federated Recommendation Framework with Differential Privacy

Fast-adapting and privacy-preserving federated recommender system

An efficient privacy-preserving point-of-interest recommendation model based on local differential privacy

Privacy-aware Location Privacy Preference Recommendations

Private and Utility Enhanced Recommendations with Local Differential Privacy and Gaussian Mixture Model

Decentralized Collaborative Learning Framework for Next POI Recommendation

Differentially Private Online Learning for Cloud-Based Video Recommendation with Multimedia Big Data in Social Networks

PrivateRec: Differentially Private Training and Serving for Federated News Recommendation

A Secure And Efficient Framework For Privacy Preserving Social Recommendation

On the resilience of Collaborative Learning-based Recommender Systems Against Community Detection Attack

Differential privacy in collaborative filtering recommender systems: a review

When Differential Privacy Meets Randomized Perturbation: A Hybrid Approach For Privacy-Preserving Recommender System

Holistic Collaborative Privacy Framework for Users' Privacy in Social Recommender Service