Research Of User Privacy Protection Methods For Personalized Recommendation

In the fast-developing information age,users use a variety of technological products to generate extremely large and complex data.If the data can be used well,the data can be used to serve users.For example,the ubiquitous recommendation problem in life(i.e.,the recommendation of movies,friends,products)is the result of good use of data.It also gives users more convenience.However,if users want to obtain high-quality recommendation services,third-party recommendation platforms should collect a large amount of user data.The user data may be used by attackers to cause privacy leakage.Therefore,privacy protection technology should be used to process user data.However,by using this privacy protection technology,the quality of recommendations might be reduced.Users might not be able to obtain high-quality recommendations.How to provide users with high-quality recommendations while user's privacy is not leaked? It is a problem that needs to be solved.This article studies this issue and conduct experimental verification of specific recommendations.Firstly,a privacy protection method based on user data confusion is investigated.The article proposes a user data obfuscation method based on sorting and classification.The core idea of the method is to sort and classify user data according to user preferences.The obfuscation processing of user high-sensitivity data reduces the loss of user data sorting.At the same time,it not only reduces the leakage of user privacy but also ensures the practicability of user data.We use the differential privacy method to confuse a large amount of user low-sensitivity data.Through the experiment of the attack model,the effectiveness of this method in privacy protection is verified.Secondly,a privacy protection method based on encrypted sparse graph data is studied.The article proposes a differential privacy protection method based on exchange.The core idea of the method is to add false edges to the encrypted sparse graph to cover up the correct node degree and the edge of the graph.First,the sparse graph data is divided into regions according to the node degree.The previous region uses the difference between the maximum and minimum node degrees of the next region to construct the Laplacian differential privacy function.Then the number of forged edges is generated according to the function.Since the forged edge is encrypted to 0,it should not affect the practicability of the data.Meanwhile,it can ensure that the privacy leakage risk of encrypted sparse graphs is reduced.Through the experiment of statistical function attack,the privacy protection effect of this method is verified.Furthermore,these two privacy protection methods are applied to the recommended field.Then two personalized recommendation model are established.Meanwhile,the two privacy protection methods are applied to specific recommendation problems.Experiments with specific user data have been conducted to verify that the two privacy protection methods have better accuracy and higher efficiency in personalized recommendation.