Recommender Systems Based on Autoencoder and Differential Privacy

Jiahui Ren, Xian Xu, Zhihuan Yao, Huiqun Yu

Abstract: Recommender systems are widely applied in practice. However the process of recommender involves the users' sensitive and privacy information inevitably. The privacy protection of recommender systems must be taken into account. In this paper, a recommender system model based on autoencoder and differential privacy is proposed. Two methods of applying differential privacy to autoencoder are designed: input perturbation and objective function perturbation. Both theoretical analysis and experimental results show that the proposed methods, as well as related algorithms, can provide reliable privacy preservation while maintaining high prediction accuracy.

External IDs:dblp:conf/compsac/RenXYY19