Reasoning human emotional responses from large-scale social and public media

Abstract

The basic characteristics of extreme events are their infrequence and potential damages to the human–nature system. It is difficult for people to design comprehensive policies for dealing with such events due to time pressure and their limit knowledge about rare and uncertain sequential impacts. Recently, online media provides digital source of individual and public information to study collective human responses to extreme events, which can help us reduce the damages of an extreme event and improve the efficiency of disaster relief. More specifically, there are different emotional responses (e.g., anxiety and anger) to an event and its subevents during a whole event, which can be reflected in the contents of public news and social media to a certain degree. Therefore, an online computational method for extracting these contents can help us better understand human emotional states at different stages of an event, reveal underlying reasons, and improve the efficiency of event relief. Here, we first employ tweets and reports extracted from Twitter and ReliefWeb for text analysis on three distinct events. Then, we detect textual contents by sentiment lexicon to find out human emotional responses over time. Moreover, a clustering-based method is proposed to detect emotional responses to a certain episode during events based on the co-occurrences of words as used in tweets and/or articles. Taking Japanese earthquake in 2011, Haiti earthquake in 2010 and Swine influenza A (H1N1) pandemic in 2009 as case studies, we reveal the underlying reasons of distinct patterns of human emotional responses to the whole events and their episodes.