Public relations (PR) companies hire and pay cooperative and organized spammer groups to post specific content on online microblogging sites, such as Twitter, to influence public opinion or trending topics (topic hijacking). Detecting such spammer groups in Twitter is challenging because they usually disguise themselves as normal users and post many tweets with common content, or they cooperate with each other in retweeting to hijack the trending topics. In addition, the spammer groups intentionally evolve their content and behavior patterns and use distributed strategies to employ only a small fraction of spam accounts to randomly hijack topics. Thus, it is even more difficult to detect them using the traditional approach of analyzing content, account, or behavior features.

The authors leverage the retweeting relationships and present a topology-based method to detect spam groups partially distributed in different trending topics to create influence. The abnormal topics are detected by the abnormal changes of topology characteristics of the retweeting networks. To measure the abnormal hijacked topics, the subgraph ranking method is presented to recognize the similarity of retweet subgraphs among different snapshots of retweet relationships over time in each topic. Once the hijacked topics have been detected, the spammer group detection algorithm works by using label propagation of a few spammer labels and recognizing the sequence of anomalous topics that the spammer joins and the set of users joining the anomalous topic sequences. This is used in calculating a cumulative user authority list, joining the anomalous topics, and clustering them to distinguish the spammer group from normal users.

As spamming attacks in social media become rampant and more sophisticated, identifying spammer groups is a step toward more trusted social media. This paper shows one effective approach to identifying the cause of topic hijacking that can mislead people’s judgment and decisions. I recommend to the authors and other researchers to employ this method to existing datasets to further verify its effectiveness. In addition, the IT companies should provide the online service platforms to adopt the proposed approach to filter out the spammer groups.