Large-scale analysis of online social data on the long-term sentiment and content dynamics of online (mis)information

Abstract

The widespread dissemination of online misinformation poses a significant threat to both our information ecosystem and the decision-making processes of individuals and societies. While prior research has extensively examined the static characteristics of misinformation, its long-term evolution remains underexplored. This study addresses this gap by analyzing a dataset from Twitter (now “X”) comprising approximately two million tweets related to 366 fact-checked stories. Consistent with existing research, misinformation carried more negative emotions, particularly disgust and anger, compared to true information, while positive emotions were notably suppressed. Importantly, over time, misinformation became increasingly negative, a trend that was not observed for true information. Furthermore, we found a decreasing trend in positive emotions for misinformation, which was not mirrored in true information. These temporal sentiment dynamics were not attributable to differing content dynamics, as there was no evidence that the dominant narratives of false and true stories evolved differently. These findings underscore the importance of studying the temporal dynamics of online (mis)information; however, as our data was limited to fact-checked stories, it may not fully represent all information shared online. Nevertheless, these findings can (i) aid the efforts of politicians, journalists, social media providers, and the public in fostering a resilient information ecosystem, and (ii) inspire further research into areas such as the influence of source credibility, cross-platform generalizability, and underlying psychological mechanisms.