AI forecasts political news engagement from 60M tweets over 7 years
New neural network predicts political lean of news users will share on Twitter
Deep Dive
Researchers trained a neural network on over 60M tweets from politically engaged users over seven years to forecast which news articles users will engage with. The model uses past news engagements and tweet content to predict the political lean of shared news. Key findings: hyperpartisan users are more engaged; right-leaning users engage with contra-partisan sources more than left-leaning users; and topics like immigration, COVID-19, Islamophobia, and gun control are salient indicators of engagement with low quality news sources.
Key Points
- Dataset: 60M+ tweets from politically engaged users collected over 7 years, with ~10% annotated for news outlet mentions and political leaning
- Neural network forecasts political lean of news a user will engage with, blending past behavior with tweet text
- Key findings: hyperpartisans engage more; right-leaning users engage more with opposite-leaning news; immigration, COVID-19, gun control linked to low-quality sources
Why It Matters
Offers a predictive tool to detect and understand filter bubbles and misinformation spread in political news consumption