Detecting Spammer on Micro-blogs Base on Fuzzy Multi-class SVM

Micro-blog has become an important information dissemination and exchange platform in people's social lives. Massive micro-blog data contains a large number of valuable information, but the micro-blog platform appears to have a lot of spam behavior problems in recent years; behavior consistent with spammers and spam micro-blogs. The spam not only affects the impact of micro-blog's data mining and decision analysis, but also seriously affects the healthy development of micro-blog platform and user experience. In this paper, a new spammer detection method based on fuzzy multi-class support vector machines (FMCSVM) is proposed in micro-blog, it combines the SVM multi-class classifier with the fuzzy mathematics theory in spammer detection. Current researches on micro-blog spammers is to analyze the characteristics of the global spammers, so that the strength of these analyses is not enough, and these researches lack the feature analysis for a certain type spammer. As a result, this will enable the spammer to escape the spam detection system. In this paper, we divide spammers into three categories by analyzing the features of micro-blog spammers, and then construct one-versus-rest SVM multi-class classifier. The fuzzy clustering method is used to deal with the mixed samples generated by the multi class classifier, and the combination classifier is obtained, which improves the detection accuracy.