Research On Personalized Differential Private Data Publication In Social Network

On the premise of protecting the personal privacy,the release of valuable social network data is one of the most challenging issues of privacy protection.Due to the feature of robustness and rigor,differential privacy appeared recent years has been applied to protect the relationships between individuals in the social networks.This model achieves privacy protection through data distortion caused by random noise.How to improve the availability of published data under the condition of differential privacy is the problem we need to solve.As far as we know,uniform privacy budget is used for all the social network data release under differential privacy,but different individuals have different needs for privacy protection.If you set a large privacy budget to ensure data availability,the protection for some users could be inadequate;On the other hand,in order to achieve the privacy protection needs of all users,excessive protection could be imposed on some users.In addition,we found that for a particular availability function,when different tuples change,the difference also aroused in corresponding query results,and before and after the data synthesis,if those tuples that have a greater impact on the query results are changed,serious impact would be caused on the availability of the data release.Our objective is to share meaningful data under the premise that the user's privacy has been protected.Motivated by this,we propose personalized edge of differential privacy for social network data,and the users can specify their own privacy protection requirements,so that the possibility of leaking individual relationships is limited.Meanwhile,we start from the important structure property the graph k-triangle counting,analyze the reasons of the changeof the results,define the concept of important data from this,and design a new method for publishing composite graph-NoiseGraph.The core of this method is as follows:(1)A totally new method for Quasi-line Partition of spatial decomposition method is proposed,which divides the two-dimensional space ? into non-uniform regions and forms IRTree,and explores sparse and dense regions better.(2)The important data of k-triangle counting shall be partitioned into the sparse or dense area as much as possible.We have compared the relative errors of traditional differential privacy and personalized edge differential privacy,and showed the application of NoiseGraph method on the counting of k-triangle subgraphs.Finally,we verify the effectiveness of personalized edge differential privacy and synthetic graph publishing method through the experimental results.