Link Prediction For Privacy And Anonymization In Dynamic Social Networks

With the rapid development of the Internet, social networks are becoming more and more popular. People communicate and share information through social networks, social networks have become a part of people's lives, and have great potential for business development. However, the popularity of social network brings privacy disclosure issues. A lot of information are exposed on social networks, which has brought an opportunity for criminals. Therefore, the privacy and anonymization of social networks has become a hot research topic.Most of the existing privacy and anonymization methods are only focused on the static social network. However, in real life, almost all of the social networks are constantly changing with time. Therefore, this thesis proposes a kind of privacy and anonymization mechanism based on generalization and grouping which is suitable for dynamic social networks, this is, the link prediction for privacy and anonymization in dynamic social networks. The true identity of the corresponding node in a social network can be hidden, and the attacker cannot determine whether there is a link between two nodes by this mechanism. This mechanism can protect users'privacy by grouping the nodes safely and using the link prediction method. Meanwhile, the mechanism improves the existing link prediction algorithms and proposes a link prediction method based on Common-Neighbors for dynamic social networks. The proposed method considers three metrics, the time-varied weight, the change degree of common neighbor and the intimacy between common neighbors. Moreover, we redefine the common neighbors by finding them within two hops, which improves the performance of link prediction and privacy protection.In this paper, the experiments are conducted based on the DBLP data set in MATLAB. The link prediction method and the privacy and anonymization method are analyzed and evaluated according to the experiments. The experimental results show that the link prediction method in this mechanism has great advantage in the performance of link prediction. At the same time, the privacy and anonymization method in this mechanism has a greater degree of effect on the privacy protection by using the safe-class-condition and the link prediction method.