Research On Mobile Social Network Based On Differential Privacy

Currently,mobile social network closely depends on the real social relations which probably face with potential attacks.Consequently,data holders may disturb or anonymize these data,and then publish them,for the purpose of privacy protection.In view of the characteristics of mobile social network,such as sensitivity of graph structure and large scale of social data,many existing privacy protection methods are frequently disabled especially for the data usability or data security.For resolving those problems,this thesis has done the following work.Based on the differential privacy model involved its realization mechanism,combination characteristic and protection mechanism,a privacy protection approach based on clustering and noise(PBCN)which can be used in mobile social network has been proposed.In PBCN,clustering,randomization and differential privacy are combined together for data privacy protection.And meanwhile,a higher sensitivity algorithm based on neighbor degree is proposed for privacy measure.And then the analysis of data availability and performance comparisons of PBCN,Spctr Add/Del,Spctr Switch and DER method are conducted under the same privacy protection conditions.The experimental results demonstrate that PBCN achieves the more satisfactory performance compared with other methodsIn mobile social network,the intricate social relationships between users are not equal,and the sensitivity of individual relationships may directly affect the distribution of privacy and the efficiency of protection.Currently,there are many privacy protection methods for social network graph without weights,however these methods cannot be directly applied to those with weighted values(i.e.uneven sensitivity of social relations).To solve this problem,it proposes a weighted social network graph perturbation method based on differential privacy protection model,which can achieve strong protection of edge weights and graph structure.This method adds disturbance noise based on the single-source shortest path constraint model,divides the critical edges and non-critical edges according to the weights,and effectively reduces the edge relationships that need to be disturbed.Different data sets are selected for simulation experiments.Compared with the classical methods,our proposal achieves the satisfactory execution efficiency and data availability,and it is suitable for large-scale social network applications.