FMNV: A Dataset of Media-Published News Videos for Fake News Detection
February 12, 2026 · View on GitHub
This repository contains the dataset FMNV (Fake Media News Videos), as presented in the paper "FMNV: A Dataset of Media-Published News Videos for Fake News Detection" (ICIC 2025).
Unlike existing datasets that focus on user-generated content (UGC), FMNV focuses on professionally produced news videos from mainstream media outlets. It is designed to benchmark multimodal fake news detection models against high-quality, deceptive content.
📥 Download Dataset
The dataset is hosted on Baidu Netdisk.
- Link: Download FMNV Dataset
- Password:
2i94
📊 Dataset Overview
Most existing fake news video datasets consist of low-quality, user-generated clips. FMNV addresses the gap in detecting high-impact fake news disseminated by media organizations, which often features professional editing and higher perceived credibility.
Statistics
- Total Videos: 2,393
- Real News: 893 (Sourced from verified Twitter & YouTube media accounts)
- Fake News: 1,500 (Generated via LLM-assisted augmentation)
- Average Duration: ~73.8 seconds (Longer than typical short-video datasets)
Categories of Manipulation
The dataset categorizes fake news videos into four distinct types based on cross-modal inconsistency:
- Contextual Dishonesty (CD): The video title misrepresents the actual events shown in the video (semantic mismatch).
- Cherry-picked Editing (CE): Critical video segments are selectively removed to create a biased or false narrative.
- Synthetic Voiceover (SV): The original audio is replaced with AI-generated speech that contradicts or fabricates the visual context.
- Contrived Absurdity (CA): Videos that maintain surface-level consistency but present exaggerated, illogical, or "common sense" defying claims.
| Category | Count | Description |
|---|---|---|
| Real | 893 | Authentic media-published news |
| Fake (CD) | 600 | Title-Video mismatch |
| Fake (CE) | 450 | Visual information deletion |
| Fake (SV) | 300 | Audio falsification |
| Fake (CA) | 150 | Logical absurdity |
📝 Citation
If you use this dataset in your research, please cite our paper:
@inproceedings{wang2025fmnv,
title={FMNV: A Dataset of Media-Published News Videos for Fake News Detection},
author={Wang, Yihao and Qian, Zhong and Li, Peifeng},
booktitle={International Conference on Intelligent Computing (ICIC)},
pages={321--332},
year={2025},
publisher={Springer},
doi={10.1007/978-981-96-9794-6_27}
}