| Popis: |
Concerns about misinformation in the Global South are growing, and strategies to combat this problem hinge on our ability to correctly detect false and biased news. Here, we provide an approach for detecting false stories that circulate as text and often without hyperlinks. We show the usefulness of this approach by examining the case of Brazil, where false information thrives, circulating largely via text, which makes it exceptionally hard to spot using established approaches. Our text-based approach relies on a combination of texts from false stories identified by fact-checkers, supervised learning methods, natural language processing, and human review of posts. We contrast our approach with the domain-based approach that is frequently used in the developed world, as well as with Facebook’s URL approach. The results show that sharing false news by politicians is a rare event: across different approaches, less than 1% of political leaders’ social media posts contain misinformation. However, we find that our text-based approach produces a different characterization of the amount of misinformation shared by political elites in Brazil, and leads to different conclusions about who shares misinformation and the type of false content shared. Importantly, we find very little overlap in the posts detected as containing misinformation across approaches. Our results show that the text-based approach is an important complement to the dominant approach as it is more effective at detecting false news. |
