Detecting Political Misinformation in the Global South: A Text-Based Approach from Brazil

It presents a text-based approach to detect misinformation, which relies on a combination of false stories identified by fact-checkers, supervised learning methods, natural language processing, and human review. The approach is compared to existing domain-based and Facebook-URL approaches, and the results show that what is identified as misinformation is largely approach-dependent. The text-based approach captures posts containing falsehoods flagged by fact-checkers, while the dominant approaches tend to detect biased or hyper-partisan content rather than strictly false content. The study finds that political leaders rarely share misinformation, and the predictors for sharing misinformation are similar across approaches.