Exploratory Analysis on Linguistic Patterns of Fake and Real News Related to the COVID-19 Pandemic

Abstract

Since the first COVID-19 case, the world has been simultaneously dealing with a pandemic and the consequences of fake news shared about it. We believe that fake and real news have different linguistic characteristics. Therefore, this study explores the linguistic and semantical differences between fake and real COVID-19 news on social media. We use a dataset collected by Patwa et al. (2020) that contains real and fake social media posts. First, we use LDA topic modelling on the collection to extract topics regarding COVID-19. In total, the model found 21 topics with a high coherence score of 0.725. The number of fake and real news articles per topic shows which topics need more careful attention by the public.

We use VADER to analyse the differences in sentiment polarity between fake and real news. The results show that fake news is statistically significantly more negative than real news overall. Additionally, specific topics (e.g., the topics “Misinformation COVID-19” and “Donald Trump”) showed more negativity within fake news than real news. Lastly, we looked at linguistic and emotional differences between real and fake news using the English LIWC dictionary. Using the Mann-Whitney test, we show that overall fake news shows statistically significantly more anger than real news. Within topics, we found further dissimilarities between fake and real news using grouped LIWC categories.