A Blockchain-based Privacy-Preserving Recommendation Mechanism

Abstract

Recommendation system is widely used to predict users' interests and provide targeted products for them, which effectively facilitates users in the era of big data where information overload problem is prevalent. Unfortunately, massive data closely related to users' privacy is in high demand to produce more accurate predictions. In this case, the collection and transmission of such data is communication costly; to process and analyze such data is of high possibility to compromise users' privacy. In this paper, we propose a privacy-preserving recommendation mechanism based on blockchain, which well addresses these problems. Leveraging the inherent advantages of blockchain, we establish a completely distributed model mitigating the risk of privacy disclosure caused by central data storage. Moreover, we combine Inter-Planetary File System with blockchain to greatly improve the communication efficiency. We also introduce local sensitive hashing and local differential privacy into proposed mechanism to reduce the computation load and provide a strong privacy guarantee. The experimental results demonstrate that the proposed mechanism shows better performance on privacy preservation while maintaining desirable recommendation accuracy when compared with the baseline.