Characterizing the Online Discourse in Twitter: Users’ Reaction to Misinformation around COVID-19 in Twitter

During a pandemic, social media is a low-cost, accessible, and broad-reaching channel to disseminate information, however social media platforms can be hotbeds for misinformation. An analysis of misinformation around COVID-19 based on social media offers insights into what users perceive as misinformation, as well as their reactions. In order to identify the topics, sentiments, and user accounts in Twitter contributing to the spread of misinformation surrounding COVID-19, we analyze 12,000 tweets posted between February and May of 2020. We employ topic identification, network analysis, and sentiment analysis to study users’ behaviors around misinformation. We identify six topics and train a set using several machine learning and neural network models to automatically classify tweets. The experimental results indicate the predictions of our models for six categories related to COVID-19 misinformation, achieving the highest accuracy of 95%. The network analysis identifies clusters of accounts and indicates how misinformation is spread. In addition, our analysis shows that sentiment scores are strongly influenced by government measures and public speeches from government officials, and the primary drivers of discourse are news agencies, public figures, health organizations, and lay citizens. In general, our proposed approaches provide a better understanding of the posts and user accounts who lead the discussion about misinformation around COVID-19.