New cross-genre information retrieval models identify misleading citations in COVID-19 tweets

What is it about?

Computer scientists from Stony Brook University and Colorado State University studied Twitter posts related to the COVID-19 pandemic, where the post cited a news article. Starting with 46.86 million tweets, the researchers have developed supervised learning models to detect factual claims in tweets. The studies presented in this paper show that over 43% of the tweets were not factual. Moreover, a significant fraction of tweets containing check-worthy claims -- 27.5% of the annotated sample (which corresponds to at least 1% of the entire test corpus) -- contained claims that appear to be true because they cite a news article, but the citation is misleading (i.e., the news article doesn't actually support the claim made in the tweet, even though that is the appearance presented at first glance to a casual reader).

Featured Image

Photo by Bud Helisson on Unsplash

Photo by Bud Helisson on Unsplash

Why is it important?

Since the beginning of COVID-19, copious information about the pandemic has been shared on social media. Unlike prior studies on misinformation, this research identifies whether there is reliable support across two distinct genres: social media and news articles. The findings are an important reminder that pandemic-related information should not always be taken at face value – even in the presence of a cited source – without carefully verifying those sources.

Perspectives