|
Download README.md from YRC10/FSD: direct link, hf CLI and curl.
- Browser
- Download file 2.03 kB
-
https://huggingface.co/datasets/YRC10/FSD/resolve/main/README.md
- Command line
-
hf download hf://datasets/YRC10/FSD/README.md
-
curl -L -o README.md https://huggingface.co/datasets/YRC10/FSD/resolve/main/README.md
2.03 kB
| license: cc-by-4.0 | |
| language: | |
| - en | |
| tags: | |
| - false-information | |
| - hurricane | |
| - wildfire | |
| - climate | |
| - disaster | |
| - social-media | |
| - multimodal | |
| - information-integrity | |
| size_categories: | |
| - n<1K | |
| # FSD: False-Info Severity Detection | |
| Dataset for the paper "[Beyond Truth Discovery: A Two-Stage Framework to Assess the Severity of False Claim during Disasters](https://arxiv.org/pdf/2608.20983)". This dataset contains false claims extracted by an agent from disaster-related Reddit posts. The two CSV files cover hurricanes (`hurricane_gt.csv`) and wildfires (`wildfire_gt.csv`). Each row represents one extracted claim, rather than a full post. | |
| | Column | Description | | |
| | --- | --- | | |
| | `claim_id` | Claim identifier. | | |
| | `false_claim` | Agent-extracted false claim. | | |
| | `believe_mv` | Believability label from majority voting by three independent annotators: `1` = low, `2` = high. | | |
| | `harm_mv` | Harmfulness label from majority voting by three independent annotators: `1` = low, `2` = high. | | |
| | `severity` | Combination of `believe_mv` and `harm_mv`, written in that order: `11`, `12`, `21`, or `22`. | | |
| **Believability** measures how likely a claim is to be perceived as credible. | |
| **Harmfulness** measures its potential consequences *if believed*. | |
| **Severity** is derived from these two labels; it is not a separate annotation. | |
| ## Citation | |
| @inproceedings{yao2026fsd, | |
| author = {Yao, Ruichen and Dasari, Tejna and Baispay, Gulshat and Zaurbek, Aizhan and Liu, Yifan and Liu, Yaokun and Li, Zelin and Wang, Dong}, | |
| title = {Beyond Truth Discovery: A Two-Stage Framework to Assess the Severity of False Claim during Disasters}, | |
| year = {2026}, | |
| isbn = {9798400728945}, | |
| url = {https://doi.org/10.1145/3834580.3838746}, | |
| doi = {10.1145/3834580.3838746}, | |
| booktitle = {Proceedings of the 2026 ACM Conference on Human-AI Complementarity and Alignment}, | |
| pages = {154–167}, | |
| numpages = {14}, | |
| } | |
| Dataset DOI: [10.57967/hf/10877](https://doi.org/10.57967/hf/10877) |