Learning to Live with COVID-19: Informative Fictions of TikTok Misinformation and Multimodal Video Analysis

Abstract

The spread of misinformation has historically been attributed to emotions, thinking styles, biases, and predispositions, but only a few studies have explored the conditions influencing its prevalence. The Theory of Informative Fictions (TIF) addresses this gap by presenting propositions that predict the conditions under which misinformation is tolerated and promoted. Building on the literature on TIF and deep learning, we uncover how property messages and character messages differ in veracity and explore the relationship between visual misinformation and user engagement. By constructing a short video dataset Tikcron ( N = 42,201) and a multimodal video analysis framework KILL, we classify TikTok videos as misinformation or not, and property messages or character messages. Our results indicate that character messages are more likely to convey misinformation than property messages, and character messages with misinformation are more likely to get tolerated and promoted by social media users than property messages with misinformation. This study extends the current methodological advancement of image-as-data to misinformation videos and proposes a multimodal video analysis framework to develop communication-centered theories. The broader practical implications of this study on the detection, countering, and governance of visual misinformation are also discussed.