wgyhhh/RealTimeNews-2025

Towards Real-Time Fake News Detection under Evidence Scarcity

Guangyu Wei*, Ke Han*, Yueming Lyu†, Yu Luo, Yue Jiang, Caifeng Shan, Nicu Sebe

(*Contribute equally, †Corresponding author)

👀 About RealTimeNews-2025

Conventional fake news datasets often comprise news that is several years old. Such instances are grounded in a wealth of post-hoc evidence, including public discussions, official statements, and scientific articles. To advance research on real-time fake news detection, we introduce a new benchmark, RealTimeNews-25, consisting of 3,487 news articles collected between June 2024 and September 2025. The dataset covers recent and rapidly evolving events characterized by limited supporting evidence, providing a challenging and timely benchmark for evaluating model robustness in real-world, time-sensitive scenarios.

⚙️ Dataset Format

The dataset is structured as follows:

├── data
    ├── news
        └── news.json
    ├── imgs
        ├── 0.png
        ├── 1.jpg
        ├── 2.png
        └── ... # {id}.jpg/png/webp

❤️ Citation

Please cite the paper as follows if you use the RealTimeNews-2025:

@misc{wei2025realtimefakenewsdetection,
      title={Towards Real-Time Fake News Detection under Evidence Scarcity}, 
      author={Guangyu Wei and Ke Han and Yueming Lyu and Yu Luo and Yue Jiang and Caifeng Shan and Nicu Sebe},
      year={2025},
      eprint={2510.11277},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
}