Spam Detection On Microblogs

Microblog has become an important platform of information propagation and sharing in social life. Massive microblog data contain much valuable information, which is likely to have practical values for government and enterprises to make decisions. For example, by utilizing microblog information, enterprises can have a comprehensive understanding on market and thereby improve their marketing strategies, while government can make better analysis on public sentiment.However, recently a great number of spammers (e.g. zombie fans) and spam microblogs (e.g. advertising information) have emerged on microblogging platforms. These spam information not only affects the effectiveness of data mining and decision analysis, but also impacts the healthy development of microblogging platforms as well as user experiences. Therefore, detecting and filtering spams is important to microblog data analysis and mining.In this paper, we focus on two issues in microblog spam detection, which are spammer detection and spam microblog detection. Spammers refer to zombie fans and advertising users, and spam microblogs refer to those microblogs whose contents are not related with specific topics. In summary, the main work and constrobutions of this paper can be summarized as follows:(1) Detection of spammersBased on a thorough analysis on microblog users’characteristics, we found that zombie fans and advertising users have different behavior. Therefore, we propose different methods to detect zombie fans and advertising users. For the detection of zombie fans, we propose a method that detects spammers with the trust/spam scores of users with the context of social relationship. For the detection of advertising spammers, we propose to use duplicated posts to detect potential spammers. The experimental results on real data sets suggest the effectiveness of our method.(2) Topic-based spam microblog detectionIn the microblogs that are related with a specific topic, there exist some microblogs whose content are not related to the current topic. These junk microblogs not only affect the browsing experiences of users, but also introduce noisy data in many kinds of work such as topic analysis. To detect microblog spams, we propose a method that considers the textual and other features of microblogs as well as the credibility of users. The experimental results on real data sets suggest the effectiveness of our method.