Caragea et al. Identifying Informative Messages in Disaster Events Short Paper – Social Media Studies Proceedings of the ISCRAM 2016 Conference – Rio de Janeiro, Brazil, May 2016 Tapia, Antunes, Bañuls, Moore and Porto de Albuquerque, eds. Identifying Informative Messages in Disaster Events using Convolutional Neural Networks Cornelia Caragea Computer Science and Engineering, University of North Texas, Denton, TX ccaragea@unt.edu Adrian Silvescu AS Research, Sunny Vale, CA silvescu@gmail.com Andrea H. Tapia Information Sciences and Technology, Pennsylvania State University, University Park, PA atapia@ist.psu.edu ABSTRACT Social media is a vital source of information during any major event, especially natural disasters. Data produced through social networking sites is seen as ubiquitous, rapid and accessible, and it is believed to empower average citizens to become more situationally aware during disasters and coordinate to help themselves. However, with the exponential increase in the volume of social media data, so comes the increase in data that are irrelevant to a disaster, thus, diminishing peoples’ ability to find the information that they need in order to organize relief efforts, find help, and potentially save lives. In this paper, we present an approach to identifying informative messages in social media streams during disaster events. Our approach is based on Convolutional Neural Networks and shows significant improvement in performance over models that use the “bag of words” and n-grams as features on several datasets of messages from flooding events. Keywords Informative tweets classification; disaster events; Convolutional Neural Networks. INTRODUCTION Much has been written concerning the value of using micro-blogging data from crowds of non-professional participants during disasters. Data produced through micro-blogging platforms, e.g., Twitter, is seen as ubiquitous, rapid and accessible (Vieweg, 2010), and it is believed to empower average citizens to become more situationally aware during disasters and coordinate to help themselves (Palen, Vieweg, and Anderson, 2010). Starbird, Palen, Hughes, and Vieweg (2010) assert that bystanders “on the ground are uniquely positioned to share information that may not yet be available elsewhere in the information space…and may have knowledge about geographic or cultural features of the affected area that could be useful to those responding from outside the area.” Despite the evidence of strong value to those experiencing a disaster and those seeking information concerning the disaster, there has been very little uptake of message data by large-scale, disaster response organizations (Tapia and Moore, 2014). Real-time message data being contributed by those affected by a disaster has not been incorporated into established mechanisms for organizational decision-making (Tapia, Moore, and Johnson 2013). Response organizations operate in conditions of extreme uncertainty. The uncertainty has many sources: the sporadic nature of emergencies, the lack of warning associated with some forms of emergencies, and the wide array of responders who may or may not respond to any one emergency. This uncertainty increases the need for appropriate information, which could make substantial improvements in the response process. We believe that data directly contributed by citizens and data scraped from disaster bystanders have a positive potential to give responders more accurate and timely information than it is possible by traditional information gathering methods. Still, information quality and use in any area of disaster response remains to be a challenge.