Reproduce “TAGFN: A Text-Attributed Graph Dataset for Fake News Detection in the Age of LLMs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kay Liu, Yuwei Han, Haoyan Xu, Henry Peng Zou, Yue Zhao, Philip S. Yu
Researchers introduce TAGFN, a large-scale, real-world text-attributed graph dataset for detecting fake news as outliers, integrating raw textual content with propagation graph structures. This resource provides a benchmark to evaluate both traditional graph learning and LLM-based methods for misinformation detection.
Key insights
- In-context learning (ICL) and Chain-of-Thought reasoning significantly enhance LLM performance for fake news detection, with one-shot examples yielding considerable accuracy improvements for smaller models.
- Supervised Graph Neural Networks (GNNs) utilizing LLM-generated embeddings consistently outperform direct LLM prompting methods across all datasets, highlighting the enduring value of abundant supervision.
- Larger LLMs demonstrate strong zero-shot capabilities, with some advanced models matching or exceeding supervised GNNs on specific subsets, and the full text-attributed graph structure (news, user posts, and graph) is essential for optimal detection performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.