arnaiztech/llms-mental-health-crisis-benchmark

Dataset Card for Between Help and Harm - Crisis Benchmark

Dataset Summary

This dataset repo contains the benchmark-side artifacts prepared for Hugging Face from the paper Between Help and Harm: An Evaluation of Mental Health Crisis Handling by LLMs, published in JMIR Mental Health.

If you use this dataset, please cite the paper. The citation is included below, the arXiv version is available at https://arxiv.org/abs/2509.24857, and the final DOI is allocated as https://doi.org/10.2196/88435.

It is organized around the example-level crisis-classification task:

• examples/: canonical validation and test conversations with deterministic example_id
• annotations/human_raw/: individual expert labels for the validation split
• annotations/human_merged/: merged human crisis labels for the validation split
• annotations/llm_raw/: raw LLM labeling runs aligned to the canonical examples
• annotations/llm_merged/: merged LLM labels aligned to the canonical examples

The files in this repo were exported from the original repository without modifying the original data/ directory. Source file names and row indices are preserved for auditability.

Repository Structure

.
├── README.md
├── examples/
│   ├── validation.parquet
│   └── test.parquet
├── annotations/
│   ├── human_raw/
│   ├── human_merged/
│   ├── llm_raw/
│   └── llm_merged/
├── metadata/
│   ├── export_summary.json
│   ├── normalization.json
│   └── provenance.json
└── checksums/
    └── sha256_manifest.json

Included Configurations

• examples: canonical example table with validation and test splits
• human_raw_labels: one row per expert label on the validation set
• human_merged_labels: merged expert labels on the validation set
• llm_raw_labels: one row per LLM label run
• llm_merged_labels: merged LLM labels per model and split

Row Counts

• examples/validation.parquet: 206 rows
• examples/test.parquet: 2,046 rows
• human_raw_labels: 4 files x 206 rows
• human_merged_labels: 206 rows
• llm_raw_labels: validation has 9 run files x 206 rows; test has 3 run files x 2,046 rows
• llm_merged_labels: validation has 1 merged file for gpt-4o-mini; test has 1 merged file for gpt-4o-mini

Schema Notes

Key columns include:

• example_id: deterministic export ID, stable within this HF export
• inputs: the original conversation turns as a list of strings
• inputs_joined: a flattened helper representation for easier inspection
• label_raw: original source label
• label: normalized label
• source_file and source_index: provenance back to the source JSON

Normalization choices are documented in metadata/normalization.json.

Label Taxonomy

The normalized label space used in this export is:

• no_crisis
• anxiety_crisis
• suicidal_ideation
• self_harm
• substance_abuse_or_withdrawal
• risk_taking_behaviors
• violent_thoughts

This export preserves the source labels and adds normalized forms where the original repository used spelling variants such as self-harm or risk_taking_behaviours.

Intended Uses

This repo is intended for:

• crisis-category classification benchmarking
• agreement analysis across human and LLM labelers
• reproducible linking to downstream response-generation and evaluation tables

It is not intended to be used as clinical guidance or as a standalone decision system for real-world mental-health triage.

Loading Examples

from datasets import load_dataset

examples = load_dataset("your-username/llms-mental-health-crisis-benchmark", "examples")
human_raw = load_dataset("your-username/llms-mental-health-crisis-benchmark", "human_raw_labels")
llm_raw = load_dataset("your-username/llms-mental-health-crisis-benchmark", "llm_raw_labels")

Provenance

• exported from the original repository JSON files
• joins validated against the original inputs field
• integrity metadata stored in metadata/provenance.json
• per-file digests stored in checksums/sha256_manifest.json

Citation

@article{arnaiz2026between,
  author = {Arnaiz-Rodriguez, Adrian and Baidal, M. and Derner, E. and Annable, J. L. and Ball, M. and Ince, M. and Perez Vallejos, E. and Oliver, N.},
  title = {Between Help and Harm: An Evaluation of Mental Health Crisis Handling by {LLMs}},
  journal = {JMIR Mental Health},
  year = {2026},
  volume = {forthcoming},
  pages = {88435},
  doi = {10.2196/88435},
  url = {https://arxiv.org/abs/2509.24857},
  note = {In press}
}

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International license (cc-by-nc-4.0).