A Privacy Protection Method Of Intimate Degree Division For Recommendation Service

In recent years,with the rapid development of the Internet and the popularity of mobile devices,Internet users are facing more and more serious problem of information overload.As the most effective solution to the problem of information overload,recommendation service can not only help users quickly locate the goods they are interested in in in massive data,but also bring huge business value to enterprises.However,with the continuous upgrading of attackers' attack means,more and more privacy leakage incidents of recommendation services are exposed in the public's view,which greatly damages the interests of users.Therefore,integration with privacy protection has become the development trend of recommendation service.At present,differential privacy is the most widely used privacy protection method in recommendation service.Compared with traditional privacy protection methods,differential privacy can provide the highest intensity of privacy protection.However,at the same time,it will destroy the data availability and affect the accuracy of recommendation service.In order to improve the accuracy of recommendation on the basis of satisfying differential privacy,this thesis proposes the idea of socialized differential privacy based on intimacy partition combined with the social relations in social networks,and proposes the corresponding intimacy partition differential privacy protectors according to the algorithm process of neighborhood and matrix decomposition,which are two mainstream recommendation services suitable for different scenarios Law.In the neighborhood oriented privacy protection method,firstly,core users are extracted by a variety of core user extraction strategies,and then privacy protection is added to the solution process of core user score matrix,user similarity matrix and user average score.In the privacy preserving method of affinity partition for matrix factorization,social differential privacy is added to the iterative process of alternating least square and random gradient descent matrix factorization algorithms respectively,so as to reduce the damage to data availability on the basis of privacy preserving.Finally,the proposed privacy preserving methods are applied to real datasets.The experimental results show that the proposed privacy preserving methods can improve the accuracy of recommendation services to a certain extent.