Spammer Group Detection Base On Graph Structure And Multi-characteristics

With the rapid development of online social network represented by Twitter and micro-blog,the network spammers use the platforms to spread rumors and deceive the public,which have seriously affected the social stability and the development of social networks.How to quickly and effectively identify these spammers plays an important role in safeguarding the network order and creating a good network environment.The traditional research on the identification of spammers has made staged progress,the research mainly through content features,user features,environmental characteristics and comprehensive features to identify these spammers.However,the spammers gradually formed spammer groups,which had a greater impact on the entire network,many researchers have started taking the spammer group as the research object.They found the network spammer based on the traditional methods,and then used clustering and community division algorithms to excavate the network spammer group.Although part of group members can be effectively excavated,but more and more spammers hide their own unusual behaviors such as content and behavior,which make the individual tend to the normal users.Although some members of the spammer group can be effectively excavated,more and more spammers made themselves tending to be normal by hiding their own unusual behaviors such as content and behavior,which leads to the accuracy of the existing network spammer group recognition method is not high.Considering that the members of the spammer group often collective action for the same garget,the overall characteristics of spammer group could not be covered.Therefore,this paper will study the network spammer group as a whole and graphically analyze the relationship among the members of the spammer and use the multiple features of the network spammer group for further testing.The paper proposes a method of spammer group identification based on graph structure and multi-features.The main identification process is as follows: First,Establish a user relationship graph for participating users under each hype blog and use the improved maximal frequent subgraph algorithm not only excavated the user groups who participate in multiple hype blog together,but also found the relationship between users.Second,taking into account the high similarity of the members of the network spammer group,users who do not have obvious similarities with other users in the user group are regarded as outliers and use outlier algorithm to filter.In this paper,the user group which is excavated by frequent subgraph algorithm and outlier algorithm is defined as the suspected network spammer groups.Thirdly,this paper analyzes the differences between the network spammer groups and the normal user groups from the structural features,content features and temporal characteristics,trains a network spammer groups classifier based on the C4.5 decision tree and uses this classifier to further determine the suspected network spammer groups and ultimately get the network spammer groups.The paper carries out an experimental analysis on the data set of Sina,and the experimental results confirmed the validity and the feasibility of the method proposed in this paper.At the same time,the experimental results show that the accuracy of the proposed method is higher.