COVID-19 Misinformation Spread: A Multi-dataset Analysis of Engagement, Propagation, and Sentiment Polarity

Abstract

COVID-related misinformation is rapidly spreading throughout social media platforms and the internet at large, posing risks not only to public health but also to political and social stability. The academic community has to a certain degree converged on the consensus that misinformation spreads faster, deeper, and more broadly than regular information, and that misinformation posts tend to express more negative sentiments and use less tentative language. Since the vast majority of prior studies focus on English datasets and Western-centric social media platforms, my paper analyzes whether their conclusions about misinformation are generalizable to other languages and platforms. Through analyses of the engagement levels, propagation patterns, textual and sentimental content, and structural characteristics of true and false posts from seven datasets, I find that while results from my English datasets are largely consistent with existing literature, the same is not true for my non-English datasets, which generally disagree with, or by some measures even directly contradict, conclusions from prior research. In particular, true posts in non-English datasets typically generate more engagement than false posts, presenting a stark contrast to one of the most well-known findings about English misinformation propagation.