The Method To Identify Spammers In Microblog

With the continuing development of Internet technology, typical online social network vehicles such as Facebook and Twitter have emerged as important tools for interpersonal communications. And in China Microblog has become the most popular online social network. While there is a huge number of spammers who spread evil information or execute vicious practices via Microblog, they also pose serious threats to the Microblog environment as well as to the privacy and asset security of other normal users. Given such, it is of great practical significance to conduct researches on anti-spammer measures, including researches on spammer identification technology.Current researches on spammer identification technology focus on two major aspects, i.e., malicious content and its statistical characteristics analysis based, and malicious user behavioral characteristics and user relations oriented. Researchers, working at one aspect or at both combined, have achieved remarkable findings. However, the current methods can only use two classification method to simply distinguish the malicious users and normal users.According to the previous experience, commencing from studies on Microblog as well as its malicious behaviors, these spammers are classified into more systematic categories based on their behavioral strategies in carrying out vicious practices. The thesis establishes a game model between normal users and spammers based on the game theory. Quantitative calculations depending on the game model are also made on Microblog users’carefulness degree, which enhances the identification between normal users and spammers by adopting user behavioral characteristics. Finally, the thesis proposes to use the confidence-based ants random walk algorithm (CARW) to identify spammers.The thesis revolves around the following issues:To begin with, we classify these spammers into more systematic categories based on their behavioral strategies in carrying out vicious practices to deal with the current situation which the malicious user classification is not clear. We acquire spammer samples as well as associated data through more diverse ways, after which these spammers are reclassified into more systematic categories based on observations and studies on their behavioral strategies in carrying out vicious practices. Given such, the thesis makes a quantitative and comparative analysis of each type of users’behavior characteristics.Secondly, in response to malicious users’adaptive changes, the thesis sets up the game model between normal users and spammers based on the game theory, calculations depending on the game model are also made on users’carefulness degree. The carefulness degree is applied to adjust users’behavior features. The introduction of carefulness degree effectively gets rid of the interference by spammers pretending to be normal users on ordinary spammer identification method. It is proven by experiments in the thesis that carefulness degree drawn from the game model makes a significant improvement on the performance of identification algorithm. The improvement ratio almost reach 5%.Finally, to change the present situation which current methods can’t multi-classify the spammers, the thesis uses the CARW to identify each type of spammers with features adjusted by carefulness degree as one of the parameters. Such semi-supervised classification brings in multiclass categorization of Microblog users, enabling correct spammer identification and classification. Experiment in this thesis proves that the method can not only have an efficient identification on the spammers but also sort them out to their own categories. The accuracy of sorting spammers out to their own categories reaches 50% and the accuracy of identifying spammers is closed to 90%.