Privacy Preserving Bloom Recommender System

Recommender Systems depend on massive amount of user data to provide accurate results. Such dependency creates a threat to individual user privacy. In this paper, a differential privacy based method is proposed to prevent the privacy attack. The Recommendation system performs predictions based on the private aggregated data from multiple clients instead of individual data. This ensures a high privacy guarantee for clients and also provides an effective recommendation. The private aggregated data from the client is collected with Local Differential Privacy (LDP). Using LDP, every client perturbs the data in their device and sends noisy data to the server. Then the Recommendation system aggregates perturbed data from all clients and computes the Recommendation. The proposed method is evaluated with a benchmarked dataset, and the results are evaluated using Precision@k, Recall@k, and nDCG@k. The experimental results confirm that it shows improved accuracy on the MovieLens 100k dataset over the existing private algorithms.