Design Of De-anonymization And Identification Algorithms In Social Networks

Nowadays,the application of online social networks is more and more extensive.These social networks have provided great convenience for people to contact and live.In order to use the services provided by the network,users have to expose their sensitive information to the network.Social network data is widely shared,forwarded and distributed to third parties,which raises the risk of a series of privacy breaches of users’ sensitive data.Therefore,data privacy issues in social networks are increasingly attracting the attention of researchers.Although the data must be anonymized before being released,such as by using the method of pseudonyms,data cleansing,and data perturbation,the attacker can still obtain the identity of user based on the collected auxiliary information.From the perspective of attackers,the collected auxiliary information can be regarded as prior knowledge to attack social networks.Researchers just used the structural features of graph to recover users’ information in social networks,ignoring the impact of user attribute information on the de-anonymization algorithm.In order to overcome this shortcoming,this paper considers the attribute similarity when calculating the total similarity between two nodes except the structure feature,which helps construct more perfect description for user’s profile.The proposed algorithm considers the influence of anonymization on user matching and sets threshold to improve the accuracy of the deanonymization.After transforming the de-anonymization problem into node matching problem of graphs,the computation complexity of algorithm can be reduced by cutting down the number of nodes to be matched at each time.According to the power-law distribution character of node degrees,the algorithm starts from a node with largest degree,which can reduce the times of comparing.At the same time,we use the method of spectral partitioning to divide the social network graph into disjoint subgraphs,which can be effectively applied to large-scale social network,and the algorithm can be effectively processed in parallel.In the following work,the user’s identity information can be further merged and collected,which can achieve more effective attack on user privacy by studying the problem of user identification across social networks.Because of the difference of structure and users’ profile between few networks,it is difficult to identify users of different networks by utilizing single feature.So,for the problem of user identification across social networks,the comparation methods based on network-attribute are proposed.Through the prior seed collected,a bipartite is constructed firstly,and then the matching between two social networks can be obtained.We can see from the data collected that almost all social networks have the attribute of ‘username’.So,the username has become an indispensable element in calculating similarity.At the same time,the algorithm designed for user identification across social networks can reduce the complexity of computation efficiently.And the evaluation verified the effectiveness of the algorithm.