README.md

---
license: cc-by-4.0
language:
- en
tags:
- false-information
- hurricane
- wildfire
- climate
- disaster
- social-media
- Reddit
- 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)