Abstract

False information is an ongoing challenge to global health crises that influences public perception and weakens emergency response efforts. This study investigates how false information is structured and framed during pandemics by using COVID-19 pandemic as a case study. Focusing on COVID-19 related Twitter posts, our study filters tweets that possibly contain false information and groups them into conversations that are analyzed through two of the main topic modeling techniques: Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorization (NMF). The results of NMF, which was found to discover more coherent and interpretable topics, were considered for theme identification and framing analysis. We were able to identify eight dominant themes that shaped the entire COVID-19 narratives. These narratives were highly politicized, but also included themes like severity, virus origin, potential treatments, health measures and global responses. The framing analysis showed that linguistic characteristics across false information often included emotionally charged words and evolved through different political and social contexts. Frames like blaming, resistance and conspiracy were recurring across the identified themes, indicating mixed feelings that amplified the spread of false information and challenged public health efforts in combating the pandemic. By combining topic modeling with manual interpretation, this research presents a novel approach to understanding the context of false information and its dynamics during times of health crises. Our findings contribute to the studies of false information and crisis communication research by showing how narratives are framed and how they evolve over time.
Keywords: False information, topic modeling, framing, COVID-19