Dinh, L., & Parulian, N. (2020). COVID‐19 pandemic and information diffusion analysis on Twitter. Proceedings of the Association for Information Science and Technology, 57(1), e252. https://asistdl.onlinelibrary.wiley.com/doi/full/10.1002/pra2.252?casa_token=QzpH0XtpmkAAAAAA%3ABiTf4V4hwwT8C3b5eR_WOm7EeNOnhIeFvwV-ZLVkslvYfBTnV-qwYUzhVqyoHEWPdL0X9hlluz7W0xcF 

University of Michigan - Ann Arbor

- Researchers studied the similarities and differences between 3 models (simulated SIR model, observed SIR model based on Covid-19 cases, and observed information cascades on Twitter) to compare how Covid-19 information is spread 
- Want to measure similarities and differences between Twitter and individual diffusion
- SIR is a compartmental model used to study how information is passed on and spread through social networks 

Research questions
- What are the diffusion patterns of COVID-19 virus spread, based on SIRsim and SIRemp?
- What are the diffusion patterns of information cascades on Twitter (INFOcas), with respect to retweets, quote tweets, and replies?
- What are the major differences in diffusion patterns between SIRsim, SIRemp, and INFOcas?

COVID‐19 pandemic and information diffusion analysis on Twitter

SIRsim
- Ran simulation on NetLogo for 88 days from December 17, 2019 to March 14, 2020 that accounted for 7.7 billion people 
- Divided population into 1,540 agents, each containing 5 million people
- Agents were simulated to travel to 36 major cities and the first agent (patient 0) started in Wuhan and started out as S
- 3 nodes – Susceptible, Infected, Removed – indicated whether agents have come into contact with patient 0 or other infected agents through driving and flying
- Simulation repeated over 100 times

SIRemp
- Used Johns Hopkins Center for Systems Science and Engineering (JHU CSSE) repository for cumulative case counts of over 185 countries from January 22 to March 14, 2020
- Used same assumptions for SIR model – S (all members of world population and thus excluded), I (confirmed cases), and R (deaths cases) 

INFOcas
- 675,228 tweets collected from December 31, 2019 to March 14, 2020 that had at least one of the hashtags #coronavirus, #covid19, #ncov
- Information cascades created for retweets, quote tweets (forward with additional information), and replies (comments) 
- SIR model – S (excluded since it represents all tweets on Twitter), I (interactions), R (no interactions during the period), new information

COVID‐19 pandemic and information diffusion analysis on Twitter methdology

What are the diffusion patterns of COVID-19 virus spread, based on SIRsim and SIRemp?
- SIMsim shows only 5% of Infected and 0.19% of Removed cases compared to world population
- Infection and deaths increase logarithmically 
- Differences in simulated and actual models found – SIMemp distribution is relatively flat compared to increasing SIRsim distribution 
- Due to not considering preventative measures in models

What are the diffusion patterns of information cascades on Twitter (INFOcas), with respect to retweets, quote tweets, and replies?
- All 3 cascades displayed linear-log distributions, specifically power law decays
- Retweets have fastest decays of 4 hours; quote tweets 3 days; reply 2 days
- While retweets are most numerous, the length of interaction with original tweets is much shorter than the interactions between quote tweets and replies

What are the major differences in diffusion patterns between SIRsim, SIRemp, and INFOcas?
- Infected cascades: strong positive correlation between SIRsim and INFOcas retweets and quote tweets; weak correlation between SIRemp and all 3 INFOcas cascades 
- Removed cascades: strongest correlations between each of the INFOcas cascades
- COVID-19 tweets get retweeted quickly, then quoted and replied

Learn Before

Related