Research On The Socail Network Attributed Graphs Via Personalized Differential Privacy Protection Algorithm

Social network has become an important data source for research in sociology and business models.However,data analysis and data mining directly on social network data can seriously threaten users' personal privacy information.Differential privacy has become a privacy protection model for social network data published due to its robustness and strict theoretical definition.Differential privacy protects the privacy of the individual by introducing random noise into the original data.The existing differential privacy for social network graphs data published method is mainly focused on graph synthesis.But the current data release method privacy budget is set by data owner,without adequate consideration of the differences in privacy requirements between individual users.And it published a topology data of social network data,which does not combine the independent attribute information and attribute information of the individual user with the correlation of edge information.According to the above problem,our paper under the condition of considering the user privacy requirements,research the influence both social network attribute graph node properties and the correlation of edge information,so we propose a social network attributes graphs algorithm under personalized differential privacy.The main research is as follows:First,We propose a personalized differential privacy social network attributes graphs algorithm(PAGA),PAGA is according to the users' privacy requirements sets to calculate the node privacy budget and the edge privacy budget,and based on the privacy budged propose an sample algorithm PEA and PNA,which are realizing the balance of personalized differential privacy attributes graphs between users' privacy requirements and total privacy budget.Second,We propose a graph synthesis algorithm AGMA,AGMA includes ANA algorithm,MKMS algorithm and CMA algorithm.ANA is proposed for the independent attribute information between users,according to the node properties do not match the data type,ANA will make the data partitioning in the data set,it also divide and calculate the probability according to the node attribute distribution query function,at last,it will sample the node according to the probability distribution.Third,In view of the correlation of edge to edge and node to edge and edge to node,we propose MKMS algorithm and CMA algorithm,MKMS algorithm using mixed kronecker product graph model to get the structure of social network properties,and compute the probability of the edge to edge information,it sample the edges according to the edge probability matrix.Edges Modified algorithm(CMA)using attributes edge correlation and its related concepts to sample edged set.Using accept/reject probability secondary sampling was carried out on the edge.This method is better to restore the topology and information of the social network graphs.Finally,through two real social network data sets,the PAGA algorithm and AGMA algorithm proposed in our paper will execute for experimental comparison,the PEA algorithm and the PNA algorithm(from PAGA algorithm)and AGMA algorithm are verified they validity and usability through the experimental results.