Optimized-FedRec: Optimized Privacy-Preserving Federated Recommender System with Elliptic Curve Cryptography based Homomorphic Encryption and Local Differential Privacy

Privacy leakage in distributed recommender system is an inevitable issue which threats the user privacy. The amalgamation of recommender system and federated learning reconstruct the crossing field “Federated Recommender System” (FedRec) where the gradient parameters from users are collected to train the recommendation model instead of raw user data. Nevertheless, federated recommender system required to dealing with issues including privacy, security and communication overhead. Furthermore, distributed system randomly sharing the user data with third party hence personal information of the user can be exposed. To resolve the described research gap, this paper proposed a novel approach using Elliptic Curve Cryptography (ECC) based fully homomorphism and Local Differential Privacy (LDP). Laplace distribution technique is exploited in LDP which randomly inject extra noise to the gradient for data perturbation to attain well-grained privatization. To ensure anonymity and integrity during communication, hash based message authentication code is incorporated with ECC. The efficiency of the proposed elliptic curve-fully homomorphism demonstrated by calculating the encryption time, decryption time, computational and communication cost. Movielens 100K and Movielens 1M are the evaluation datasets and Root Mean Square Error (RMSE), Mean Absolute Error (MAE) are the evaluation metrics of proposed federated crypto privacy-preserving with LDP model. Experimental results and comparative analysis demonstrated the outperformance of proposed model with existing privacy-preserving techniques.