Characterizing and Countering Communal Microblogs During Disaster Events

Abstract

The huge amount of tweets posted during a disaster event includes information about the present situation as well as the emotions/opinions of the masses. While looking through these tweets, we realized that a large amount of communal tweets, i.e., abusive posts targeting specific religious/racial groups are posted even during natural disasters-this paper focuses on such category of tweets, which is in sharp contrast to most of the prior research concentrating on extracting situational information. Considering the potentially adverse effects of communal tweets during disasters, in this paper, we develop a classifier to distinguish communal tweets from noncommunal ones, which performs significantly better than existing approaches. We also characterize the communal tweets posted during five recent disaster events, and the users who posted such tweets. Interestingly, we find that a large proportion of communal tweets are posted by popular users (having tens of thousands of followers), most of whom are related to media and politics. Further, users posting communal tweets form strong connected groups in the social network. As a result, the reach of communal tweets is much higher than noncommunal tweets. We also propose an event-independent classifier to automatically identify anticommunal tweets and also indicate a way to counter communal tweets, by utilizing such anticommunal tweets posted by some users during disaster events. Finally, we develop a real-time service to automatically collect tweets related to a disaster event and identify communal and anticommunal tweets from that set. We believe that such a system is really helpful for government and local monitoring agencies to take appropriate decisions like filtering or promoting some particular contents.