ABSTRACT
This paper mines social media conversations collected using #sideeffects during the early phases of the Covid-19 vaccine roll out. The multi-dimensional analysis seeks to uncover post-vaccination symptoms and their intensity, latent underlying emotions, how the discourse spreads and is received, and ultimately separate the content into two groups according to their severity. The paper infers that people share their severe symptoms that cause major inconvenience and disruption using exaggerated emotions, and the follower networks of these authors are stronger. Tweets outlining mild symptoms are liked and retweeted many more times, and the friends networks of these authors are stronger. Both types of tweets (severe and mild) express heavy surprise and some fear. Classification of the tweets into mild and severe groups is attempted using boosting, neural network and BERT-based natural language transformer methods. BERT significantly outperforms the former two achieving a F1-score of 0.88. Given that the Covid-19 vaccine was developed and deployed rapidly, with limited clinical data on its side effects, this research fills an important gap in gathering such information under real-life settings. The paper thus concludes with a discussion of how these findings could be leveraged by public health organizations to combat vaccine hesitancy and maximize vaccine uptake. © 2022 IEEE.