This paper introduces a robust, distance-based differential privacy (DP) framework called Privacy-Preserving Collaborative Filtering (PPCF) to protect user data in recommender systems (RSs) while maintaining high-quality recommendations. The proposed model integrates self-organizing map (SOM) clustering with collaborative filtering (CF) and differential privacy to enhance both privacy and personalization. A local adjustment strategy is applied to balance data availability with privacy, supported by aggregation, generalization, ϵ -budget management, and Laplace noise injection to mitigate inference attacks. Social network signals were also incorporated to further refine user preference modeling. The system was evaluated using the MovieLens dataset, in which the trade-off between privacy and accuracy was examined under various ϵ values. Key metrics such as the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were used to assess the prediction performance. The experimental results demonstrate that PPCF consistently outperforms traditional non-private models and several existing differentially private methods. The model shows improved clustering stability and recommendation accuracy even at lower ϵ levels, though results confirm that small variations in ϵ can notably affect clustering quality. These findings underscore the importance of privacy-utility tuning for practical deployment. Overall, the proposed PPCF approach offers a privacy-resilient, accurate, and scalable solution for collaborative filtering in environments utilize sensitive data.

Top