Social Network SYBIL Group Detection Based On User Attributes And Behavior Charactristics

With the rapid development of information technology,online social networking platform has become an indispensable part of people's lives.Social network for people to build social networks and social relations platform while also facing increasingly serious Sybil attack.Sybil user by manipulating Internet voting,publish a lot of advertising and information generated more negative impact on the normal use of social networks.In order to detect hidden in ordinary user of Sybil users,researchers have proposed many features based on attributes and behavioral characteristics and detection algorithms have achieved good detection effect,but with Sybil users to enhance the effect of camouflage,just from the user own property can not be on the user's authenticity as more reliable evaluation,and these methods ignore the importance of different attributes for user evaluation of the authenticity of the difference.VoteTrust algorithm solves the problem can not be accurately detected due to the invasion of Sybil user's user community caused but the method can not cope on concern-forward type of social network users mistakenly real concern higher rate of problems.This paper proposes a weighted support vector machine based group detection model Sybil,Sybil detection compared with the existing model,the paper user attributes when selecting a machine learning method to solve not only consider the characteristics of the user's own property,and increase the user all property features followers can more effectively detect Sybil imitate real user attributes and can reduce false positives for users similar to Sybil's real only when users choose their own property.Experimental data based on user access to online social networks showed that the properties of paper selected feature vector by support vector machine classifier was Sybil group compared to existing detection model has high accuracy.Then based on user attributes feature is proposed based on Euclidean distance credibility solve the model.To reduce the complexity of the algorithm,so that the principle of concise,first Sybil user attribute value range of statistics and find the center of the property value range,then the user attribute value Sybil computing users with the possibility that the value of the distance,the model uses ROC curve of the experimental results were evaluated and found property credibility can better achieve the Sybil and the real user to distinguish.Finally,bump function model to optimize the properties of credibility,reduce the credibility of groups suspected Sybil value while increasing the credibility of suspected real user value,to provide better reliability parameters behavior model.On the basis of the properties on the credibility of the credibility we proposed a property value as an important parameter model Sybil group VoteTrust improved user real calculation.To VoteTrust method to be forwarded to the attention characterized by social networks,this article will focus on the behavior of the one-way equivalent to send a friend request,and the introduction of property characteristics to the preliminary evaluation of the credibility of the user,such as micro-blog for online social networks,Sybil groups in the interest of individual objects,objects are concerned,or the number of fans is different,according to these attributes,you can determine whether the individual in a social network whether Sybil users.Then according to real users hardly send Sybil users concerned requests fidelity propagation weights,get Sybil classification.This paper collected from Sina Weibo newer experimental data,and manually marked Sybil nodes analytical modeling,and recently published a number of detection algorithms for comparison,results show that detection algorithm based on behavior characteristic attributes have a higher confidence detection rates and lower false positive rate.