Shahi, G. K., Dirkson, A., & Majchrzak, T. A. (2021). An exploratory study of covid-19 misinformation on twitter. Online Social Networks and Media, 22, 100104. https://www.sciencedirect.com/science/article/pii/S2468696420300458

University of California, Berkeley

This study looked at the propagation, authors, and content of COVID-19 misinformation on Twitter from January to July of 2020 through exploratory analysis. 

An exploratory study of COVID-19 misinformation on Twitter 

In this study, researchers collected two datasets. The first dataset consists of false/partially false tweets from fact-checking websites. The second dataset consists of a random sample of COVID-19 related tweets. 

To obtain the first dataset, they first collected 7623 fact-checked news articles from Snopes and Poynter. Then, they used Beautifulsoup to crawl the articles and anchortags to obtain the tweet ID's, collecting 3053 Tweet ID's. Tweepy was then used to fetch the Tweets. 

To obtain the second dataset, they randomly extracted 1000 COVID-19 related tweets per day between the months of January and July through Twitter API and Twitter4J, getting 92095 tweets from Twitter API and 71000 tweets from Twitter4J. 

Tweets were categorized into "false," "partially false," "true," or others based on fact-checks. 

An exploratory study of COVID-19 misinformation on Twitter Data Collection

An exploratory study of COVID-19 misinformation on Twitter

Bot detection: A bot detection API was used on each twitter account. 

Type of account: TwiRole was used to classify each account as a brand or normal user using profile name, picture, latest tweet, and account description. 

Popularity of account: The popularity of an account was determined by looking at the followers count, verification status, favorites count, and account age. 

Propogation: This equation was used: $P_s = \frac{ \sum_{n}^{d} -1^{rc}}{N_d}$ where $P_s$ is the propagation speed, rc is the retweet count per day, and $N_d$ is the total number of days. 

Content analysis: Most common hastags and emojis, distinctive terms, and language indicating emotions or psychological factors were noted to analyze the content of the tweets. 



An exploratory study of COVID-19 misinformation on Twitter Methods

- misinformation (false category) gets more likes than the partially false category
- spread of misinformation peaked between March 16th and April 23rd
- false tweets spread faster than partially false tweets (propagation speed)
- misinformation often discredits information circulating on social media
- completely false information, compared to general COVID-19 tweets, often talks about the government
- partially false information is more concerned with transmission and mortality rates
- Iauthors posting misinformation are driven by affiliations to others significantly more often than authors of COVID-19 tweets in general and significantly less often by rewards or achievements

Learn Before

Related