Research On Integrated Recommendation Algorithm With Differential Privacy

With the rapid development of Internet technology,recommendation system has been widely used in e-commerce websites.According to user's historical purchase record,recommendation system can recommend products that users may be interested in,so that users can experience better recommendation services,and then increase online sales.Recommendation service relies on users' historical rating information.Once the recommendation system is attacked,users' private information may be disclosed.In order to protect privacy of users,it is necessary to provide privacy protection for the user data in the recommendation system.The traditional privacy protection technology is difficult to resist the background knowledge attack,but the differential privacy technology is proposed in recent years that can solve this problem.The differential privacy technology is to protect privacy of users by adding controllable noise to sensitive information of users.This thesis mainly studies two kinds of integrated recommendation schemes for differential privacy protection.The main work is as follows:(1)An integrated recommendation scheme(DP-IR,where IR represents the integration of SDCF algorithm and MF algorithm)based on differential privacy protection that is designed.The DP-IR scheme is mainly implemented by three steps,namely pretreatment,neighborhood based similarity calculation and matrix factorization target perturbation algorithm.The DP-IR scheme integrates the user collaborative filtering algorithm and matrix factorization algorithm by constructing the balance coefficient,which can make full use of the local and global information of the rating matrix.At the same time,it takes into account the consistence based on collaborative filtering algorithm and the sparse set based on matrix factorization.Theoretically,it is proved that DP-IR algorithm satisfies ?-differential privacy and can achieve project-level privacy protection.And in the real data set,this thesis does many experiments to verify that DP-IR algorithm can protect users' data privacy and security,and can achieve high-quality recommendation services.(2)An integrated recommendation scheme based on personalized differential privacy(PDP-IR,where IR represents the integration of PPCF algorithm and MF algorithm)that is designed.PDP-IR scheme applies the personalized differential privacy mechanism to the integrated recommendation algorithm,which is mainly implemented by PPCF algorithm,sampling mechanism and target perturbation algorithm in matrix Factorization.Considering that each user that has different requirements for the privacy level of various projects,PDP-IR scheme can provide users with personalized privacy requirements,provide users with project-level privacy protection,and provide users with high-quality recommendation services.According to the definition of personalized differential privacy,it is proved theoretically that PDP-IR scheme satisfies personalized differential privacy,so as to ensure the privacy security of the whole scheme.Moreover,multiple experiments are carried out on the real data set to verify the practicability of PDP-IR scheme.