| During natural disasters or crises, users on social media tend to easily believe contents of postings related to the events, and retweet the postings, hoping that the postings will be reached by many other users. Unfortunately, there are malicious users who understand the tendency and post misinformation such as spam and fake messages with expecting wider propagation. To resolve the problem, in this paper we conduct a case study of the 2013 Moore Tornado and Hurricane Sandy. Concretely, we (i) understand behaviors of these malicious users; (ii) analyze properties of spam, fake and legitimate messages; (iii) propose at and hierarchical classication approaches; and (iv) detect both fake and spam messages with even distinguishing between them. Our experimental results show that our proposed approaches identify spam and fake messages with 96.43% accuracy and 0.961 F-measure. |
