Meta-Learning and Targeted Differential Privacy to Improve the Accuracy-Privacy Trade-off in Recommendations

2026-04-29

AI Summary

This work addresses the significant degradation in recommendation accuracy caused by uniform noise injection in conventional differential privacy approaches for privacy-preserving recommender systems. To mitigate this issue, the authors propose a targeted differential privacy mechanism that selectively perturbs only those user data instances exhibiting high privacy risk, identified through a precise privacy risk assessment. Furthermore, meta-learning is integrated to enhance the model’s robustness to injected noise. By focusing perturbations on sensitive samples rather than applying them uniformly across all data, the method achieves lower empirical privacy risk while substantially improving recommendation accuracy. Experimental results demonstrate that this approach outperforms existing baselines employing uniform or full-data differential privacy, yielding a superior trade-off between privacy preservation and utility.

Abstract

Balancing differential privacy (DP) with recommendation accuracy is a key challenge in privacy-preserving recommender systems, since DP-noise degrades accuracy. We address this trade-off at both the data and model levels. At the data level, we apply DP only to the most stereotypical user data likely to reveal sensitive attributes, such as gender or age, to reduce unnecessary perturbation; we refer to this as targeted DP. At the model level, we use meta-learning to improve robustness to remaining DP-noise. This achieves a better trade-off between accuracy and privacy than standard approaches: Meta-learning improves accuracy and targeted DP leads to lower empirical privacy risk compared to uniformly applied DP and full DP baselines. Overall, our findings show that selectively applying DP at the data level together with meta-learning at the model level can effectively balance recommendation accuracy and user privacy.

Problem

• differential privacy
• recommendation accuracy
• privacy-accuracy trade-off
• privacy-preserving recommender systems

Innovation

• targeted differential privacy
• meta-learning
• privacy-accuracy trade-off
• recommender systems
• noise robustness