Semi-supervised self-training for COVID-19 misinformation detection: analyzing Twitter data and alternative news media on Norwegian Twitter

Abstract This paper investigates the dissemination of COVID-19 misinformation on Twitter within the context of the Norwegian media landscape, characterized by high levels of trust in the media, yet experiencing an increasing influence of alternative news sources. Using a semi-supervised self-training approach for text classification, a dataset of 426,262 tweets is analyzed, identifying approximately 5.11% as misinformation. The study reveals that misinformation tweets receive heightened engagement, particularly in retweets, and originate predominantly from a small group of users. Furthermore, while misinformation tweets are more likely to link to alternative news media sites, these sites represent only a minor fraction of the overall links shared. The analysis highlights distinct temporal patterns, with misinformation activity spiking during significant events such as the arrival of COVID-19 vaccines in Norway and the emergence of the Omicron variant. This research underscores the complexity of misinformation dynamics in a high-trust media environment and emphasizes the need for effective strategies to combat misinformation, particularly from alternative news media that challenge conventional narratives while often propagating falsehoods. Overall, the findings contribute valuable insights into the interplay between social media, alternative news sources, and misinformation dissemination during a global pandemic.