Misinformation proliferated on social media undermines efforts to respond to public health threats, with adverse effects on the adoption of preventative health behaviors. To mitigate ill effects of misinformation, it is important to understand how misinformation is received and responded to by social media users. To this end, we introduce a novel measurement of the controversy of a social media post, based on the variability in the stance taken in response to it. For the initial task of stance detection, we used in-context learning with Llama 3.1, with accuracy rates of 85.3% and 71.8% on the COVID-19 stance detection task across two annotated datasets. We estimated controversy by computing the entropy of the predicted stances, to evaluate the hypothesis that tweets containing misinformation would be more controversial. Our methods provide a new perspective from which to observe how information is received in social media, with the potential to inform the design of public health interventions to mitigate misinformation.
Li et al. (Thu,) studied this question.