ABSTRACT
During the COVID-19 pandemic, different groups had different perceptions of how dangerous the coronavirus was. This difference in behavior was intensified by the large amount of misinformation shared across social media. This work presents an analysis aimed at understanding the extent to which people perceived risk at different levels, and at uncovering the relationship between these differences and the spread of misinformation. In particular, we focus on Brazil, because it is well-known that its Ministry of Health has sponsored campaigns that raised suspicious regarding the effectiveness of the vaccines. To achieve this goal, we gathered tweets written in Portuguese related to the COVID-19 and analyzed their psycholinguistic traits. Among those traits, we found 'Anxiety' to be a good proxy for risk perception. We validate this choice by showing that, at moments of high (resp. low) infection rates in the world, the Anxiety score was higher (resp. lower). We grouped users into 'low' and 'high' risk perception based on the users' anxiety score, and analyzed the relation of each group with the spread of misinformation. Our results show that Twitter users with a lower perceived risk were more inclined to share fake news and harmful information, while the group with a higher level of anxiety tends to share more scientifically-backed information. This is an important step towards helping minimize the spread of false and harmful health information around the internet. © 2022 IEEE.