Research On A Recommender System Based On Differential Privacy

With the widespread development of Internet technology,there are more network data,and it is more difficult for users to obtain information.However,recommendation system can help users to dig out their interests from a wide range of services,so it is very popular in the industry and academia.However,in the field of recommendation algorithm,the accuracy of the model depends heavily on the data quality.Completely data can better promote model optimization,but it brings great data security problems.Some of the information is related to the user privacy,and users don't want to disclose it.Therefore,research on privacy protection methods is imperative.Compared with other ways,the advantage of differential privacy is that it provides a complete privacy protection framework.Therefore,in order to promote the development of recommendation system,this paper studies the combination of recommendation and differential privacy.The main work and research results are as follows:First of all,this paper studies the typical attack and defense methods in the field of data security,and deeply considers the security risks of recommendation system.Because the existing research mainly focuses on traditional machine learning,this paper innovatively proposes the topic of privacy protection of deep recommendation model.Secondly,based on the important parameters of deep learning,this paper integrated and innovated the original PNN model,and used the grid search method to determine the optimal model configuration,so as to obtain the best results in the current data set.Then,based on the characteristics of the recommendation system,the corresponding dp-pnn scheme is implemented by adding Laplace noise and gradient truncation,and the effectiveness of the scheme is proved.Finally,for the definition of differential privacy,this paper proposes a method to verify the effectiveness of differential protection,so that the algorithm can be verified by experiments to meet the degree of differential privacy protection.For the proposed scheme,this paper not only analyzes the data security from the theoretical perspective,and proves that the algorithm meets the privacy requirements,but also balances the accuracy of the algorithm and the privacy guarantee from the experimental perspective,and proves the effectiveness of data protection.This research has a certain practical value.