An Approach for Detecting Spammer Invasion Events with Poisson Process

The spammer, referring to the user who sends large amount of unrelated information to recipients without permission in a short period, brings a serious threat to social network security. Interestingly, we find that the change of network structure can be helpful for the identification of spammer invasion events. In the paper, a method which aggregates fifteen kinds of network structure indicators together is proposed to detect spammer invasion events in time series based on Poisson process. Tested on the dataset of Tagged.com, our proposed approach outperforms other state-of-the-art methods in precision, recall and F1-score. Also, the result validates that the key indicators for detecting spammer invasion events are comprised of ABC, SHC, MD. The method presented will be refined and extended to cases of detecting spammers dynamically.