[Paper Review] Ginger Cannot Cure Cancer: Battling Fake Health News with a Comprehensive Data Repository

Enyan Dai, Yiwei Sun+1 more|arXiv (Cornell University)|Jan 27, 2020

Misinformation and Its Impacts15 citations

TL;DR

This paper introduces FakeHealth, the first comprehensive, publicly available repository of fake health news featuring richly annotated news content, detailed review explanations, social media engagements, and user network data. It enables explainable and knowledge-enhanced detection of health misinformation, with exploratory analyses validating dataset quality and identifying key challenges in the domain.

ABSTRACT

Nowadays, Internet is a primary source of attaining health information. Massive fake health news which is spreading over the Internet, has become a severe threat to public health. Numerous studies and research works have been done in fake news detection domain, however, few of them are designed to cope with the challenges in health news. For instance, the development of explainable is required for fake health news detection. To mitigate these problems, we construct a comprehensive repository, FakeHealth, which includes news contents with rich features, news reviews with detailed explanations, social engagements and a user-user social network. Moreover, exploratory analyses are conducted to understand the characteristics of the datasets, analyze useful patterns and validate the quality of the datasets for health fake news detection. We also discuss the novel and potential future research directions for the health fake news detection.

Motivation & Objective

• Address the critical gap in comprehensive, multi-faceted datasets for fake health news detection.
• Provide a rich, annotated dataset that supports explainable and knowledge-based detection approaches.
• Enable researchers to study propagation patterns and user behavior in health misinformation spread.
• Facilitate the development of robust, interpretable models for detecting fake health news using diverse data sources.
• Support future research in health misinformation by releasing a high-quality, scalable, and privacy-compliant dataset.

Proposed method

• Constructed two datasets—HealthStory and HealthRelease—containing news content with source publishers, images, and metadata.
• Collected and structured 500k tweets, 29k replies, 14k retweets, and 27k user profiles with timelines and friend lists from Twitter.
• Annotated each news item with detailed reviews using ten standardized evaluation criteria (C1–C8, S9–S10, R9–R10) to assess credibility and bias.
• Preserved user privacy by storing only IDs of social engagements and user networks, while providing API access to full data via Twitter’s public API.
• Organized data into four main folders: contents, reviews, engagements, and user_network, each with structured JSON files for programmatic access.
• Enabled reproducibility and scalability by version-controlling the repository and maintaining it for ongoing usability.

Experimental results

Research questions

• RQ1 What are the key characteristics and propagation patterns of fake health news on social media?
• RQ2 How do social media users engage with real versus fake health news, and what role do bots and normal users play?
• RQ3 Which criteria in news reviews most strongly differentiate fake from real health news?
• RQ4 Can explainable detection models be effectively trained using the annotated review criteria and social context?
• RQ5 How can knowledge graphs derived from review explanations improve the interpretability and accuracy of fake news detection?

Key findings

• The FakeHealth repository contains 500k tweets, 29k replies, 14k retweets, and 27k user profiles with timelines and friend lists, enabling large-scale social network analysis.
• News reviews cover ten evaluation criteria, including evidence quality, conflict of interest disclosure, and sensational language, providing a foundation for explainable detection.
• Exploratory analysis revealed that both normal users and bots contribute to the spread of fake and real health news, with significant differences in propagation patterns.
• The dataset supports detection models that leverage publisher credibility, visual content, and social context, improving performance beyond text-only approaches.
• The inclusion of structured review annotations allows for training models that identify specific flaws in health news, such as misrepresentation of risk or lack of funding disclosure.
• The dataset’s design enables future research in knowledge-based detection by extracting background knowledge from reviews to build interpretable health knowledge graphs.