Research On Collaborative Filtering Recommendation Algorithm Based On Privacy Protection

With the continuous development of the Internet and the progress of e-commerce,product recommendation and privacy protection become more and more important.The recommendation system aims to solve the problem of information overload by providing accurate recommendation items to users.Collaborative filtering is considered to be one of the most widely used recommendation algorithms,which provides the similar items or similar users recommendation for users in online environment.In addition,collaborative filtering algorithm can be used with trusted network,so as to deliver the rating values from users and suggestions of trusted network to the querying users.On the other hand,users may have privacy concerns and are reluctant to submit the personal information required by the system,for example,items ratings,which makes the recommendation system unable to provide enough accurate recommendations.Therefore,it is very important to protect the privacy of users while considering the accuracy of recommendation.This essay focuses on the collaborative filtering recommendation algorithm based on privacy protection by studying the existing recommendation algorithm and privacy protection technology,and has made some innovations.The main contributions of this essay are followed.(1)This essay proposes a recommendation method based on collaborative filtering algorithm and user trusted network.(2)This essay proposes a privacy protection method applied on the server to provide additional level of protection for its users.This method is based on an improved role-based access control model to prevent unauthorized access to internal user data.In addition,randomization disturbance technique is used for user similarity matrix before recommendation generation.(3)In view of medical recommendation scenarios,a collaborative filtering recommendation model based on multi-party random masking and polynomial aggregation is proposed,which is suitable for any distributed medical data.The model generation is divided into two stages,offline model generation and online prediction generation.Three kinds of privacy protection protocols are proposed,and privacy security analysis is carried out for each protocol.The security,accuracy,coverage and performance of the scheme are analyzed on real medical data set.The experimental results show that the scheme can protect the privacy of data owners,and the privacy protection measures adopted do not affect the accuracy of the prediction results.Based on the offline computing cost and online computing cost,the comparison and analysis of the proposed scheme with other related schemes show that the performance of the proposed scheme is significantly improved.