[Paper Review] Challenges in Combating COVID-19 Infodemic -- Data, Tools, and Ethics

Kaize Ding, Kai Shu+3 more|arXiv (Cornell University)|May 27, 2020

Misinformation and Its Impacts8 references4 citations

TL;DR

This paper identifies three core challenges—data quality, tool effectiveness, and ethical trade-offs—in combating the COVID-19 infodemic using AI. It proposes a framework leveraging social media data, weak supervision, and explainable detection models to identify misinformation, while emphasizing privacy-preserving design and multi-disciplinary collaboration to ensure ethical, robust, and scalable solutions during public health crises.

ABSTRACT

While the COVID-19 pandemic continues its global devastation, numerous accompanying challenges emerge. One important challenge we face is to efficiently and effectively use recently gathered data and find computational tools to combat the COVID-19 infodemic, a typical information overloading problem. Novel coronavirus presents many questions without ready answers; its uncertainty and our eagerness in search of solutions offer a fertile environment for infodemic. It is thus necessary to combat the infodemic and make a concerted effort to confront COVID-19 and mitigate its negative impact in all walks of life when saving lives and maintaining normal orders during trying times. In this position paper of combating the COVID-19 infodemic, we illustrate its need by providing real-world examples of rampant conspiracy theories, misinformation, and various types of scams that take advantage of human kindness, fear, and ignorance. We present three key challenges in this fight against the COVID-19 infodemic where researchers and practitioners instinctively want to contribute and help. We demonstrate that these three challenges can and will be effectively addressed by collective wisdom, crowdsourcing, and collaborative research.

Motivation & Objective

• To address the urgent need to combat the spread of misinformation, conspiracy theories, and scams during the COVID-19 pandemic.
• To identify key challenges in data availability, computational tool limitations, and ethical risks in deploying AI for infodemic mitigation.
• To promote collaborative, multi-disciplinary research that ensures fairness, transparency, and long-term sustainability in AI-driven disinformation detection.
• To develop explainable and privacy-aware systems that maintain user trust while effectively detecting false content.
• To prevent long-term harm from overreliance on surveillance or biased models by embedding ethical considerations into the design process.

Proposed method

• Leverages weak supervision from social media engagement patterns and fake news detection scores to train models with limited labeled data.
• Uses publisher-newspiece-consumer relationships to model credibility, incorporating sociological insights on media bias and user trust.
• Applies explainable AI techniques to mine user comments and highlight specific text segments that contribute to fake news predictions.
• Employs a website matcher to cross-reference URLs with known disinformation sources from NewsGuard for improved detection accuracy.
• Designs contact-tracing systems with end-to-end encryption and peer-to-peer communication to minimize privacy risks.
• Promotes transparency and user consent in data collection to balance public health needs with civil liberties and democratic values.

Experimental results

Research questions

• RQ1 How can we effectively detect and explain misinformation about COVID-19 using limited labeled data and social media signals?
• RQ2 What are the key technical and ethical challenges in deploying AI tools to combat the infodemic at scale?
• RQ3 How can we ensure privacy and prevent misuse of health and behavioral data collected during public health emergencies?
• RQ4 In what ways can collaborative, multi-disciplinary research improve the robustness and fairness of disinformation detection systems?
• RQ5 What design principles can balance public health urgency with long-term democratic and ethical integrity in surveillance and data use?

Key findings

• The volume of coronavirus-related spam nearly tripled in one week, with nearly 3% of all global spam now related to the pandemic.
• Eight of the top 10 most engaged-with news sites in the U.S. are spreading false content about the virus, indicating high virality of misinformation.
• A significant proportion of the public—nearly 30% of Americans—believe the virus was a lab-created bioweapon, highlighting widespread conspiracy theory dissemination.
• Fake news detection models trained with weak supervision and social engagement data show improved performance in identifying disinformation without requiring extensive manual labeling.
• Explainable AI methods successfully identify specific text segments and user comments that contribute to false news predictions, increasing model transparency.
• Privacy-preserving contact-tracing systems using end-to-end encryption and decentralized data storage can reduce risks of surveillance abuse while maintaining public health utility.