Research And Application On Differential Privacy Preserving In Personalized Recommender Systems

In recent years,with the rapid development of Internet technology and the arrival of the era of big data,people's daily work and life will inadvertently generate hundreds of millions of data information through a variety of ways.Faced with such a large amount of data information,personalized recommendation system can find valuable information,which has made important contributions to the development of human society and economy.Although it contributes to the work and life of mankind,but there is a risk of revealing the user's sensitive information because of the need to collect a large amount of user information.When users find that the uploaded information may expose their privacy in the recommended analysis,more and more users are not willing to share his data or they may share some false data for recommended analysis.This is contrary to the original intention of the original system.Therefore,how to guarantee the safe of user privacy and how to provide accurate results have become a hot topic for current research.Differential privacy preserving framework is the basic environment for the implementation of the differential privacy.In this paper,the ProPer framework satisfied with PDP preserving is studied and improved.The framework is used in the distributed environment.Then,the D-ProPer difffernetial privacy preserving framework is proposed.In the framework,aimed at this problem what GS division method may cause great errors if there are outliers in the dataset,an S-GS data publishing method satisfied with ?-differential privacy is presented.Based on the original GS division method,the conception of the set of D-value to determine the outliers in the data set is introduced for the targeted division.In order to prove the superiority of the S-GS method,the method is compared with the GS method in two group experiments.Experimental results show that when a dataset has a large number of outliers,the S-GS differential privacy data publishing method has a stronger robustness and better data utility than the GS method.For the problem that data mining may lead to leakage user privacy,this study research the K-means clustering method based on differential privacy preserving.In order to improve the poor usability and robustness of clustering algorithm influenced by the sensitivity of outliers,a differential privacy A-PAM clustering algorithm is proposed based on the PAM algorithm.Make sure to achieve privacy preserving while maximizing cluster availability through loop iteration.The experimental results show that the differential privacy A-PAM clustering algorithm is implemented for privacy preserving.It also has better clustering availability and greater robustness of the algorithm.Finally,the paper puts the theory of the algorithm proposed in the paper to the practical application and designs a personalized recommendation system about film.This system provides commom mode and private recommendation mode.The user can choose the corresponding recommendation mode according to their own requirements of data privacy.In private mode,the system will be combined with the S-GS method and A-PAM clustering algorithm for rigorous screening and recommending.To some extent,it avoids the problem of user privacy leakage caused by the recommendation analysis process.