Answers are grounded in this paper and cite their sources.
Recommender Systems (RSs) output ranked lists of items, such as movies or restaurants, that users may find interesting, that users may find interesting, based on the user’s past ratings and ratings from other users. RSs increasingly incorporate Differential Privacy (DP) to protect user data, raising questions about its impact on recommendation accuracy and fairness. We conduct a comprehensive, cross-model evaluation of two DP mechanisms, Differentially Private Stochastic Gradient Descent (DPSGD) and Local Differential Privacy (LDP), applied to 8 recommender systems (Neural Collaborative Filtering (NCF), Bayesian Personalized Ranking (BPR), Singular Value Decomposition (SVD), Variational Autoencoder (VAE), Denoising Autoencoder (DAE), Light Graph Convolution Network (LightGCN), K-Nearest Neighbor (KNN), and Most Popular-based Recommender (MostPop)) on the MovieLens-1M, Yelp, Amazon Grocery and LastFM datasets. We find that stronger privacy consistently reduces utility, but not uniformly. NCF under DPSGD shows the smallest accuracy loss (under 10% at ϵ ≈ 1), whereas SVD and BPR experience larger drops. VAE and DAE are among the most privacy-sensitive architectures, with sharp declines for sparsely represented groups. In contrast, LightGCN combined with LDP maintains strong utility across datasets and achieves NDCG values above 0.8 on LastFM under privacy constraints. The impact on bias metrics is similarly heterogeneous. DPSGD generally reduces the gap between recommendations of popular and less popular items, whereas LDP preserves existing patterns. These results highlight that no single DP mechanism is uniformly superior; instead, each provides trade-offs under different privacy regimes and data conditions.
No takes yet. Share an insight, caveat, or question.
Parsarad et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: