Overview
Kai-Cheng Yang presents findings from his team's large-scale comparative study of COVID-19 misinformation on Twitter and Facebook. The talk covers data collection methodology, prevalence analysis, platform differences, and the role of verified accounts as "superspreaders" of low-credibility content.
Key Findings Presented
- Platform prevalence: Twitter shows higher prevalence of low-credibility COVID-19 content (32% of tweets) compared to Facebook (21% of posts)
- Data scale: Analysis of 53 million tweets from 12 million users on Twitter and 37 million posts from 140k pages/groups on Facebook (January–October 2020)
- Superspreaders: Small group of verified accounts responsible for disproportionate amplification (40% of retweets on Twitter, 70% of reshares on Facebook)
- Identification method: Low-credibility sources identified via Media Bias/Fact Check (MBFC) database; high-credibility sources curated from major news outlets
- Temporal patterns: Misinformation prevalence peaked during March 2020 pandemic surge, then stabilized
- Platform differences: Ranking of misinformation sources differs significantly between platforms (Spearman r = 0.57)
- Coordinated behavior: Evidence of coordinated sharing patterns among clusters of Facebook pages and Twitter accounts
- YouTube analysis: Removed/deleted videos show different popularity patterns across platforms; reuptake on alternative platforms provides insight into removed content
Methodological Contributions
- Concentration measure based on inverse normalized entropy to quantify whether low-credibility sharing is distributed or concentrated
- Cross-platform alignment strategy accounting for platform-specific mechanics (retweets vs. shares/reshares)
- Manual annotation of removed YouTube videos via archive sites and alternate video platforms
Related Paper
The full peer-reviewed results are published in: 2021-yang-covid-infodemic (Big Data & Society, 2021)