---
license: cc-by-4.0
task_categories:
- text-retrieval
language:
- en
---

# CiteGuard Dataset

CiteGuard is a benchmark and framework used in the paper [CiteGuard: Faithful Citation Attribution for LLMs via Retrieval-Augmented Validation](https://huggingface.co/papers/2510.17853). It reframes citation evaluation as a problem of citation attribution alignment, assessing whether LLM-generated citations match those a human author would include for the same text.

- **Project Page:** https://kathcym.github.io/CiteGuard_Page/
- **GitHub Repository:** https://github.com/KathCYM/CiteGuard

## Dataset Format

The dataset follows a CSV format with the following columns:

- `id`: Unique identifier for the excerpt.
- `excerpt`: The text containing the citation to be validated (e.g., using a `[CITATION]` placeholder).
- `year`: The year of the source paper.
- `source_paper_title` (optional): The title of the paper containing the excerpt.
- `target_paper_title` (optional): The gold standard title(s) for evaluation.

## Sample Usage

Using the [CiteGuard repository](https://github.com/KathCYM/CiteGuard), you can run the evaluation on a dataset file via the CLI:

```bash
python -m src.main --model_name gpt-4o --dataset DATASET.csv --result_path results/run.json
```

## Citation

If you use this dataset or the CiteGuard framework, please cite the following:

```bibtex
@misc{choi2026citeguardfaithfulcitationattribution,
      title={CiteGuard: Faithful Citation Attribution for LLMs via Retrieval-Augmented Validation}, 
      author={Yee Man Choi and Xuehang Guo and Yi R. Fung and Qingyun Wang},
      year={2026},
      eprint={2510.17853},
      archivePrefix={arXiv},
      primaryClass={cs.DL},
      url={https://arxiv.org/abs/2510.17853}, 
}
```