Recommendation System With Privacy Protection Based On Fully Homomorphic Encrpyption

With the rapid growth of total data amount on the Internet,the existence of recommendation systems has brought great convenience to society.By helping people find information or items related to their preferences,recommendation systems reduce the time that people spend finding information they need from such huge amount of data.But when people enjoy the convenience of recommendation systems,they're also concerned about the data security and privacy protection of recommendation systems.Not only the user data may be leaked in every step of the recommendation system,but also with other public data the anonymous identity can be restored from the simply processed data,which could expose users' sensitive information.It can be seen that simple data protection can not solve the privacy disclosure problem.Therefore,the research on privacy protection mechanism in recommendation systems has great practical significance.At present,the research on privacy protection of recommendation systems is mainly based on homomorphic encryption to provide data protection.This paper mainly studies the privacy protection of recommendation systems,and proposes a matrix factorization scheme based on homomorphic encryption and a privacy protection recommendation system implementation.The main contributions of this paper are as follows:(1)Matrix factorization is a common algorithm in the recommendation system,decomposing the users' rating data matrix into user and item profile vectors.Currently,executing the matrix factorization algorithm under fully homomorphic encryption results in large computation and interaction overhead in the system due to the complex vector inner product operations in the original matrix factorization algorithm.And because the ciphertext data is too large,the existing research schemes can only calculate a part of data in dataset.This kind of computational complexity appears to be of theoretical significance in the context of homomorphic ciphertext,and has no possibility of practice.This paper proposes an algorithm that combining homomorphic encryption and matrix factorization algorithm together.The new scheme can improve the performance of the recommendation system under privacy protection.The computational complexity of the proposed scheme does not increase with the size of the data set,but only relates to the number of users and items in the system.So that the training iteration of the whole data set can be completed within a certain computational cost.Our experiments show that the scheme has good performance and accuracy.When the recommendation server and encryption server complete the specified protocol,the users' privacy information is not disclosed.(2)The matrix factorization is only an algorithm for training the user and item profile in a recommendation system.To achieve a usable recommendation system,we need to design the system from a high level.Therefore,based on the matrix factorization algorithm of homomorphic encryption,we propose a usable recommendation system design.With the cooperation of several specialized modules,the system speeds up the responding of user requests.It also solves the cold boot problem of new users and new items,as well as some problems caused by key exchange and matrix factorization algorithm.Finally,a highly usable recommendation system with privacy protection that can quickly add new users and new items is implemented,which can recommend specific or popular items to users.It can quickly and accurately recommend appropriate items to users while ensuring user data privacy.