Source-linked AI summary
The COVID-19 Social Media Infodemic
Matteo Cinelli, Walter Quattrociocchi, Alessandro Galeazzi, Carlo Michele Valensise, Emanuele Brugnoli, Ana Lucia Schmidt, Paola Zola, Fabiana Zollo, Antonio Scala
TL;DR
The paper examines how information and misinformation spread during COVID-19 across different social media environments. Using comparative data from five platforms and epidemic models, it finds platform-dependent misinformation amplification, while reliable and questionable information show no significant differences in spreading patterns.
Problem
The paper addresses limited comparative evidence on how platform environments shape information and misinformation diffusion during a health emergency.
Method
The authors analyze COVID-19 content across Twitter, Instagram, YouTube, Reddit, and Gab, measuring engagement, discourse evolution, epidemic-model parameters, and questionable-source diffusion.
Results
Information spreading differs by platform: mainstream platforms are less susceptible to misinformation diffusion, but reliable and questionable information show no significant differences in spreading patterns.
Takeaways & Limitations
Platform interaction patterns and audience characteristics are associated with information and misinformation spreading, with rumors’ amplification varying across platforms.
Takeaways & Limitations
Standard epidemic models may not capture more complex social contagion, as illustrated by Instagram’s abrupt user increase and implausibly high estimated R0.
Abstract
from arXiv · showhide
We address the diffusion of information about the COVID-19 with a massive data analysis on Twitter, Instagram, YouTube, Reddit and Gab. We analyze engagement and interest in the COVID-19 topic and provide a differential assessment on the evolution of the discourse on a global scale for each platform and their users. We fit information spreading with epidemic models characterizing the basic reproduction numbers $R_0$ for each social media platform. Moreover, we characterize information spreading from questionable sources, finding different volumes of misinformation in each platform. However, information from both reliable and questionable sources do not present different spreading patterns. Finally, we provide platform-dependent numerical estimates of rumors' amplification.
Introduction
The paper frames COVID-19 as an infodemic challenge because information diffusion can affect public behavior and epidemic countermeasures. It therefore compares information spreading across five platforms and models platform-specific dynamics and misinformation susceptibility.
- Motivation: Information diffusion can influence public behavior and alter the effectiveness of government countermeasures during epidemics.The paper links misinformation to fragmented social responses and describes travel disruption following an early lockdown rumor.
- Motivation: Social-media algorithms facilitate content promotion, shaping social perceptions, narratives, policy-making, and public debate.The paper emphasizes these effects particularly for controversial issues.
- Research objective: The study addresses platform-specific information dynamics by comparatively analyzing Twitter, Instagram, YouTube, Reddit, and Gab.It contrasts this approach with prior misinformation studies focused mainly on single platforms.
- Research objective: The dataset contains more than 8 million comments and posts collected over 45 days to assess engagement, interest, and discourse evolution globally.The analysis covers COVID-19-related content across each platform.
- Methods: Epidemic models estimate each platform’s basic reproduction number R_0, with R_0 > 1 indicating the possibility of an infodemic.Here, secondary cases are users who begin posting about COVID-19.
- Findings: Mainstream platforms are less susceptible to misinformation diffusion, although reliable and questionable information show no significant difference in spreading patterns.Questionable-source spreading is characterized for all channels except Instagram.
1 Results and Discussion
Across five platforms, COVID-19 information showed platform-specific interaction and timing patterns. Epidemic modeling and source comparisons indicate that misinformation volumes and amplification vary by platform, while reliable and questionable content generally follow similar growth dynamics.
- Interaction patterns: User activity exhibits a long-tailed distribution across platforms, indicating broadly similar patterns of reactions and content consumption.The platforms nevertheless differ in engagement levels and discussed topics.
- Interaction patterns: The largest increases in COVID-19 posting occurred on January 21 for Gab, January 24 for Reddit, January 30 for Twitter, January 31 for YouTube, and February 5 for Instagram.All platforms showed a behavioral change around January 20, near the WHO’s first situation report.
- Information spreading: The study models information growth with both phenomenological EXP and standard SIR epidemic models, interpreting R_0 > 1 as the possibility of an infodemic.R_0 represents the average number of new users posting about COVID-19 generated by one already-posting user.
- Information spreading: Instagram’s abrupt author-growth jump cannot be explained by continuous epidemic models, producing an unrealistic SIR estimate of R_0 ∼10^2.The authors note that social contagion can be more complex than epidemic-like spreading captured by R_0.
- Questionable versus reliable information: Questionable and reliable posts and engagements grow with strong linear correlations, but their relative volumes and amplification differ substantially across platforms.Gab has about 70% as many questionable posts as reliable posts but about 270% more engagements, while YouTube has α ∼10%, Reddit α ∼50%, and Gab α ∼400%.
2 Conclusions
The study compares COVID-19 information dynamics across five social media platforms, finding that platform-specific interaction patterns shape spreading and misinformation susceptibility. It also estimates platform-dependent rumor amplification parameters.
- The study compares COVID-19 information spreading across Twitter, Instagram, YouTube, Reddit, and Gab during the health emergency.
- It assesses user engagement, topic interest, and discourse evolution over time on each platform.
- Epidemic models provide basic growth parameters for information spreading on each social media platform.
- Gab is identified as the environment most susceptible to misinformation diffusion, although reliable and questionable information show no significant spreading-pattern differences.
- Information spreading appears driven by platform-specific interaction paradigms and the interaction patterns of engaged user groups.
- The study concludes by providing COVID-19 rumor amplification parameters for each social media platform.
3 Materials and Methods
The study combines platform-specific data collection, text analysis, and epidemic modeling to compare COVID-19 information dynamics across social media. It also evaluates regression patterns and limitations in matching posts to source classifications.
- 3.1 Data Collection: Data were collected from Gab, Reddit, YouTube, Twitter, and Instagram using platform APIs, an archive, keyword and hashtag searches, and manual inspection.Instagram required a custom manual collection process because no API was available.
- 3.2 Text analysis: The analysis uses word embeddings and PAM clustering with cosine distance to identify COVID-19-related topics across platform corpora.Topics are assigned to contents whose dominant topic proportion exceeds 0.5.
- 3.2 Text analysis: The Skip-gram model represents words as vectors, with training quality affected by vector dimensionality and the surrounding-word window.The study uses 200-dimensional vectors and a six-word window, after cleaning English-language corpora and filtering rare or short content.
- 3.3 Epidemiological Models: Information growth is modeled with both an adjusted exponential model and the mechanistic SIR model to estimate platform-specific basic reproduction numbers R0.The exponential model uses least-squares parameter estimates with bootstrap ranges, while the SIR model represents infection and recovery rates.
- 3.3 Epidemiological Models: In the models, incidence is interpreted as the number of authors who have published posts about COVID-19, while SIR’s I + R corresponds to the same author count.For SIR, R0 = β/γ is the ratio between infection and recovery rates.
- 3.4 Regression Table of Figure 3: The linear regressions shown in Figure 3 have overall high R2 values, but source matching is sometimes weak, particularly for YouTube.Matching ability measures finding news outlets belonging to the Media Bias/Fact Check list, not merely identifying known domains.