Conversations in Crisis: A Network Analysis Approach to Emotion Exchange on Twitter

Abstract

This thesis examines the dataset “Emotional-Message Exchanges During 18 Crisis Events” curated by Kusen and Strembeck. The data consists of over 1.4 million Tweets from nearly 650,000 unique users regarding 18 crisis events (natural disasters, riots, shootings/terror attacks) that occurred in 2017 and 2018. Researchers labeled each Tweet with the dominant emotion according to Plutchik’s wheel of emotions. This thesis builds upon the work of Kusen and Strembeck by introducing Russell’s Arousal and Misery axes and by examining the temporal effect on triadic structures. It was found that Low Arousal and Low Misery emotion sub-networks exhibit more reciprocity, or conversation, than their High Arousal and High Misery counterparts. Emotion-specific reciprocity in the Shooting and Terror Events can be attributed to anger and neutral messaging almost exclusively; whereas, there is a greater range of emotions presenting emotion-specific reciprocity across the Natural Disaster and Riot events. Three regimes of the triadic frequency charts were identified (Initialization, Growth, Finalization) and used to identify trends in the triadic structure over time. While the role of the 021D triad as an initializing triad is often linked to Low Arousal and Low Misery emotions in Shooting and Terror Events, 021D as an initializing triad in Riot Events corresponds to High Misery emotions. Limitations of the research include that, while this a global dataset, the Tweets analyzed are strictly in English. Emotional expression varies culturally, meaning that some sentiment may be missed or misrepresented in the dataset. In the future, this research could be expanded to more platforms, such as TikTok, and could examine more demographic data on users (ex. age and location).