Differentially Private Cross-Silo Recommendation from Implicit Feedback

Cross-silo recommendation from implicit feedback is a key task in modern recommender systems, where user-item interaction data are distributed across multiple parties and cannot be centrally collected. Unlike explicit feedback, which provides fully observed real-valued ratings, implicit feedback is one-class and extremely sparse, recording only users' actions or inactions (e.g., clicks, visits, or bookmarks), yet it is far more prevalent in real-world applications. Such behavioral data are often highly sensitive, raising significant privacy concerns when used for collaborative model training. Although differential privacy (DP) has been widely applied to explicit feedback-based models, the resulting utility degradation makes it difficult to apply DP effectively to implicit feedback learning. In this work, we propose DPIMF, a differentially private implicit matrix factorization framework for cross-silo recommendation based on objective perturbation. To improve utility, we redesign the loss function and adopt an importance sampling scheme to reduce the noise scale required for privacy preservation. We further provide formal utility guarantees for the proposed techniques and characterize the conditions under which utility improvements are maximized. Experiments on three benchmark datasets validate our theoretical analysis and demonstrate that DPIMF achieves a better privacy-utility trade-off than state-of-the-art methods.

Online services recommend what to watch, read, or buy by learning from the clicks and views you leave behind. When several organizations—say, separate online retailers or streaming services—want to pool what they've each learned to make better recommendations, the shared model can quietly memorize and reveal what individual people did. A mathematical guarantee called differential privacy prevents this by mixing in random noise, but recommenders are unusually hard to protect: they treat every item a person didn't click as a faint "no," creating millions of signals that demand so much noise the recommendations turn to mush. We introduce DPIMF, which protects individuals without wrecking accuracy. Its key trick is to rewrite the learning problem so the noise only needs to hide each person's small, personal contribution, while the large pooled part—which no single person can change much—stays untouched. Two further ideas keep the math stable and reuse data sparingly to "amplify" privacy at almost no cost. On real-world datasets, DPIMF gives noticeably better recommendations than previous private methods, especially when privacy protection is strongest—making collaborative recommendation safer to deploy.