Summary: Using artificial intelligence to examine millions of tweets, researchers found that public attitudes about COVID-19—and attitudes about treatments and policies—spread rapidly on social media and can be more contagious than the virus itself.
Source: Northwestern University
Researchers at Northwestern Medicine report that public beliefs about COVID-19 and related treatments propagate widely on Twitter, shaping public behavior and opinion. Their study applied interpretable AI models to analyze how tweets, retweets and public events influence people’s health beliefs, and compared the impact of scientific evidence with statements made by political figures.
Key findings from the study include:
- Exposure to tweets amplifies individual biases: people are more likely to accept and further share a message the more it has been retweeted.
- Both scientific milestones (such as peer-reviewed publications) and non-scientific events (including political speeches) have comparable influence on public health belief trends observed on social media.
“During the pandemic, social media has been a dominant source of information and misinformation, shaping public perceptions of the disease, treatments and policy,” said Yuan Luo, corresponding author of the study and chief Artificial Intelligence officer at the Institute for Augmented Intelligence in Medicine at Northwestern University Feinberg School of Medicine.
Luo emphasized the practical implications: “Our analysis is intended to alert people that the information they encounter every day may be reliable or unreliable. It encourages readers to seek information supported by solid scientific evidence. We also aimed to provide guidance for scientists and healthcare professionals so they can more effectively reach specific audiences.”
How can scientists respond when politicians promote inaccurate claims?
Luo noted that statements by politicians—such as minimizing the risk of COVID-19 or overstating a treatment’s efficacy—can have an effect on public belief as strong as scientific publications. “These non-scientific messages can shape attitudes and behavior. That is a major concern,” he said. The research suggests that scientists and public health professionals need to proactively communicate evidence-based information to counteract misleading narratives.
“If scientists don’t actively disseminate accurate information, their efforts can be overwhelmed by louder, less responsible voices,” Luo added. He recommended investing in targeted public information campaigns—especially to educate the public about vaccines—to increase vaccine uptake and maximize public health benefit.
How do tweets influence individual attitudes?
“Every retweet and every trending post contributes to shaping people’s beliefs,” Luo said. “Individuals should pause and fact-check before sharing content. Many users are unaware of how strongly tweets influence their opinions and how easily biased information can spread. The retweeting mechanism creates a viral amplification effect similar to viral marketing: the more attention a post receives, the more it shapes the broader conversation.”
What makes this study different?
This work combines machine learning techniques with classical epidemiological modeling to trace how messages about COVID-19 circulated on Twitter and how those messages aligned with shifts in public belief. The researchers prioritized model interpretability so other investigators can understand and reproduce the AI’s behavior rather than treating it as an opaque “black box.”
The team mapped fluctuations in public attitude derived from tweets and then aligned those trends with prominent scientific and non-scientific events to identify likely drivers of change. “By linking attitude trends to real-world events, we offer actionable insights for public health communication,” Luo said.
How large was the dataset?
The research team, led by first author Hanyin Wang, collected COVID-19–related tweets via the Twitter API. Their dataset included 92,687,660 tweets from 8,967,986 users between January 6 and June 21, 2020. To build and validate the AI model, the team randomly selected 5,000 tweets for manual annotation. Each tweet was independently reviewed by two annotators and classified according to four core constructs of the health belief model: perceived susceptibility, perceived severity, perceived benefits and perceived barriers.
Next steps: applying AI to vaccine attitudes
Luo’s team is extending this approach to study public attitudes toward COVID-19 vaccines. By integrating machine learning and deep learning methods, researchers aim to identify specific public concerns and to design targeted outreach that maximizes vaccination impact. The team is also exploring whether social media data can reveal disparities by gender or race during and beyond the pandemic.
Other Northwestern contributors to the study include Yikuan Li, Meghan Hutch and Andrew Naidech.
Funding: The work was supported in part by grant R01LM013337 from the National Library of Medicine of the National Institutes of Health.
About this artificial intelligence research news
Source: Northwestern University
Contact: Marla Paul – Northwestern University
Image: The image is in the public domain
Original Research: The study was published in the Journal of Medical Internet Research and appeared during the week of March 1, 2021.