| In the current era of big data,the Internet industry is developing rapidly,data information is growing rapidly,and the problem of information overload is becoming more and more serious.With the massive information,it becomes more and more difficult to get the information users to need accurately and quickly.To solve this dilemma,personalized recommendation systems have emerged.As one of the mainstream technologies of personalized recommendation algorithms,the collaborative filtering recommendation algorithm has the characteristics of a good recommendation effect and a simple algorithm.It provides personalized recommendations by using the information of users’ historical behavior data to predict the items users need.However,in practical applications,collaborative filtering recommendation algorithms have problems like poor scalability,sparse scoring data,and user privacy leakage,resulting in the final recommendation results not meeting the real needs of users.In this paper,to solve the problems in collaborative filtering recommendation algorithm,we introduce clustering analysis technology and differential privacy technology to improve the recommendation accuracy while safeguarding the privacy of users,and the main research works are as follows(1)With the increase of data information processed by the collaborative filtering recommendation algorithm,the scalability of the algorithm will become poor.To address this problem,the personalized collaborative filtering recommendation algorithm based on differential privacy clustering is further proposed by introducing clustering analysis technology based on the collaborative filtering recommendation algorithm.The difference between this algorithm and other collaborative filtering recommendation algorithms based on cluster analysis is that in the stage of K-means clustering processing of user data,it can ensure the privacy of users participating in the recommendation system is unleaked,guarantee data security,and improve the recommendation performance of the algorithm.(2)User data for the K-means clustering algorithm will produce the problem of privacy leakage.To address this problem,we analyze the existing privacy budgetε allocation methods and design the Adaptive Differential Privacy K-means(ADPK-means)algorithm that adaptively allocates the privacy budget based on the clustering effect.The algorithm evaluates the effect of each iteration to generate clustering sets and adds different perturbation noise to different clustering sets,thus reducing redundant noise addition.At the same time,the algorithm addresses the problem that randomly selected centroids leading to poor clustering results and uses the average difference degree of sample points to select the initial centroids to ensure user privacy and security and improve the usability of the results.(3)The collaborative filtering recommendation algorithm uses users’ private information to make recommendations,which can produce privacy leakage.To address this problem,a differential privacy recommendation algorithm based on clustering results is designed.The clustering analysis technique is used to search for similar user groups.Differential privacy protection is applied to the resulting similar user groups to narrow the range of added noise.This method effectively solves the problem of direct adoption of differential privacy techniques,which leads to poor recommendation performance.Then,the ADPK-means algorithm is combined with the collaborative filtering algorithm to perform differential privacy clustering operations on users based on user attribute data to guarantee the privacy of users in the clustering process.At the same time,the generated clustering set uses the index mechanism to output the neighboring user sets,and the index mechanism utility function is designed to improve the accuracy of recommendation results by using the user-item attribute preference features and user-item rating features.The theoretical security analysis and experimental results of the algorithm proposed in this paper are conducted to verify the proposed methods are feasible and effective. |
