Emotion classification of social media posts for estimating people’s reactions to communicated alert messages during crises

Abstract

One of the key factors influencing how people react to and behave during a crisis is their digital or non-digital social network, and the information they receive through this network. Publicly available online social media sites make it possible for crisis management organizations to use some of these experiences as input for their decision-making. We describe a methodology for collecting a large number of relevant tweets and annotating them with emotional labels. This methodology has been used for creating a training dataset consisting of manually annotated tweets from the Sandy hurricane. Those tweets have been utilized for building machine learning classifiers able to automatically classify new tweets. Results show that a support vector machine achieves the best results with about 60% accuracy on the multi-classification problem. This classifier has been used as a basis for constructing a decision support tool where emotional trends are visualized. To evaluate the tool, it has been successfully integrated with a pan-European alerting system, and demonstrated as part of a crisis management concept during a public event involving relevant stakeholders.