Twitter Emotion Analysis in Earthquake Situations

Emotion keyword spotting approach can detect emotion well for explicit emotional contents while it obviously cannot compare to supervised learning approaches for detecting emotional contents of particular events. In this paper, we target earthquake situations in Japan as the particular events for emotion analysis because the affected people often show their states and emotions towards the situations via social networking sites. Additionally, tracking crowd emotions in the Internet during the earthquakes can help authorities to quickly decide appropriate assistance policies without paying the cost as the traditional public surveys. Our three main contributions in this paper are: a) the appropriate choice of emotions; b) the novel proposal of two classification methods for determining the earthquake related tweets and automatically identifying the emotions in Twitter; c) tracking crowd emotions during different earthquake situations, a completely new application of emotion analysis research. Our main analysis results show that Twitter users show their Fear and Anxiety right after the earthquakes occurred while Calm and Unpleasantness are not showed clearly during the small earthquakes but in the large tremor.