TrustworthinessTransparencySafetycs.SI
TAGFN: A Text-Attributed Graph Dataset for Fake News Detection in the Age of LLMs
Kay Liu, Yuwei Han, Haoyan Xu, Henry Peng Zou, Yue Zhao, Philip S. Yu
November 26, 2025
arXiv.org
Kay Liu, Yuwei Han, Haoyan Xu, Henry Peng Zou, Yue Zhao, Philip S. Yu
LLMs need better tools for detecting misinformation: The paper identifies that while Large Language Models have advanced text analysis, their application to fake news detection through graph-based methods remains "significantly underexplored," suggesting current AI systems may not be optimally designed to combat misinformation at scale.
Quality datasets are essential for trustworthy AI: The authors emphasize that "scarcity of large-scale, realistic, and well-annotated datasets" has been a major barrier to developing reliable fake news detection systems, highlighting that responsible AI development depends on having proper benchmarks to test against.
Fake news exists within networks, not in isolation: By framing fake news detection as a "graph outlier detection" problem, the paper suggests that misinformation is best understood through the connections between sources, stories, and users—not by analyzing individual articles alone, which has implications for how platforms should approach content moderation.
Both traditional and AI-based approaches matter: The dataset is designed to evaluate "both traditional and LLM-based graph outlier detection methods," indicating that newer AI systems should be compared against established techniques rather than replacing them entirely, supporting a more cautious approach to AI deployment.
Open resources accelerate responsible AI development: By making TAGFN publicly available, the authors enable independent researchers to test and improve fake news detection systems, supporting transparency and collaborative progress in building more trustworthy AI rather than concentrating these capabilities in single organizations.
Large Language Models (LLMs) have recently revolutionized machine learning on text-attributed graphs, but the application of LLMs to graph outlier detection, particularly in the context of fake news detection, remains significantly underexplored. One of the key challenges is the scarcity of large-scale, realistic, and well-annotated datasets that can serve as reliable benchmarks for outlier detection. To bridge this gap, we introduce TAGFN, a large-scale, real-world text-attributed graph dataset for outlier detection, specifically fake news detection. TAGFN enables rigorous evaluation of both traditional and LLM-based graph outlier detection methods. Furthermore, it facilitates the development of misinformation detection capabilities in LLMs through fine-tuning. We anticipate that TAGFN will be a valuable resource for the community, fostering progress in robust graph-based outlier detection and trustworthy AI. The dataset is publicly available at https://huggingface.co/datasets/kayzliu/TAGFN and our code is available at https://github.com/kayzliu/tagfn.
Preprint. Under review
Citation data from Semantic Scholar