Social Media–Detected Early Disaster Events: An E-Warning System of Emergency Medical Operation Centers Deployed in Taiwan (Preprint)

Abstract

BACKGROUND Because messages are broadcast on social media timelessly and efficiently, it is a critical disaster management survey tool. Taiwan’s Regional Emergency Medical Operation Center (REMOC), supervised by the Ministry of Health and Welfare of Taiwan (MOHW), has responded to nationwide disaster messages in real time since 2005. OBJECTIVE To construct an early warning system to reduce the time delay between the outbreak of accidents and messages received by REMOC. METHODS We developed a program to harvest online messages from numerous social media platforms. Messages containing disaster-relevant keywords, such as typhoon, flooding, and traffic accidents, were scrutinized. We applied text analytics methods and an open-source classification tool LibShortText, a support vector machine (SVM) approach, to identify disaster-related messages. REMOC’s on-duty staff used this information system to harvest disaster-related messages and cases between January 2019 and December 2019. The frequency of events and the percentage of the number of messages that were posted faster than the number of messages that were shared were calculated according to the messages shared by traditional media or posted by individual social media users. The engagement rate (ER) was applied to infer the response rate of registered messages by either traditional media or social media users. RESULTS The system filtered 3022 disaster-related messages from 66 588 total messages. The majority of the messages were related to earthquakes (23%), followed by traffic accidents (17%), fires (12%), and floods (3%). A total of 702 messages were related to 142 earthquake episodes. Of them, 35 messages were shared by social media users regarding 18 earthquakes, and 667 messages from traditional media concerned 124 earthquakes. Social media users posted 100% of the messages faster than traditional media reported them (the “early dissemination efficacy” by social media; 124/124). A total of 367 messages regarded 251 fires; social media users shared 122 messages regarding 82 fires, and traditional media posted 245 messages about 169 fires. The early dissemination efficacy of social media for fires was 0.57 hours. A total of 117 messages from 32 floods were harvested, 55 messages about 16 floods were shared by social media users, and 62 messages for 16 floods were posted by traditional media users; the early dissemination efficacy of social media regarding floods was 0.65 hours. CONCLUSIONS Our results indicate that social media users responded to disaster news faster than did traditional media, especially regarding earthquakes, fires, and floods. Of all disasters, earthquakes and floods had the greatest response rate from social media users. The combined usage of REMOCs and social media crawler system could considerably improve the preparation for and the response to natural disasters, particularly earthquakes and floods.