Collaborative Filtering Method Based On Differential Privacy And Application In Location Recommendation

With the rapid development of the Internet,IoT and Cloud Computing,the amount of information contained is increasingly outstanding in cyberspace.As an effective information filtering method,recommendation system can extract the personalized preferences and make accurate recommendation by analyzing the potential information of users.However,the recommendation system with large amount of user data for collaborative filtering will raise serious risk privacy of individuals.The issue of privacy protection has become the main bottleneck of recommendation system in open data environment.How to protect privacy information from disclosure has become one of the greatest challenges to recommender systems.Differential privacy has risen as a new solution for privacy protection with strong privacy guarantees preventing attacks with relevant background information.To address the problem of data security and availability in recommendation systems,this paper studies the differential privacy technology,data publishing technology and application based on location application scenario.The content of this paper is mainly as follows.Firstly,we propose a data release model of differential privacy protection based on compressed sensing.The measurement matrix is constructed to effectively alleviate the data sparse by using compressive sensing.Furthermore,noise disturbance method is used to add noise to the measurement vector satisfying the difference privacy constraint.After that,the reconstruction algorithm is used to recover the reconstructed data matrix.Secondly,a collaborative filtering recommendation algorithm is proposed to meet the needs of the privacy protection in data opening scene.By establishing the low dimensional feature matrix of the potential eigenvectors,the high-dimensional and sparsity are alleviated.Then the objective perturbation function is used to minimize the parameters in the squared error training model by stochastic gradient descent algorithm.In the process of optimization,a matrix decomposition model is obtained by adding the noise to the characteristic matrix to satisfy the difference privacy constraints.At last,the validity prediction of the differential privacy matrix decomposition algorithm is evaluated on the MovieLens and Netflix datasets.Finally,by applying a collaborative filtering recommendation algorithm to the differential privacy protection on location recommendation.Then combining DBSCAN clustering algorithm to divide the interest area,the introduction of differential privacy in the iterative process reduces the risk of user privacy leakage and improves the availability of data.The user's check-in frequency is integrated into the user similarity calculation,and the user's location is explored to share privacy preferences and predict the behavior of users.Through the experimental verification on the Gowalla datasets,the proposed method has good recommendation effect and privacy protection level.