README.md

---
license: mit
task_categories:
  - question-answering
language:
  - vi
tags:
  - medical
pretty_name: f
---

# Dataset Card for Vietnamese Mental Health Question Answering (VMHQA) Dataset

## Table of Contents

- [Introduction](#introduction)
- [Dataset Details](#dataset-details)
- [Dataset Overview](#dataset-overview)
- [Dataset Structure](#dataset-structure)
- [Dataset Creation](#dataset-creation)
- [Dataset Demo](#dataset-demo)
- [Dataset Access](#dataset-access)

---

## Introduction

**VMHQA** is a pioneering Vietnamese Multiple-Choice Question Answering (MCQA) dataset designed specifically for the mental health domain. Comprising 10,000 high-quality question-answer pairs across 1,166 subjects, this dataset addresses critical gaps in mental health resources for Vietnamese-speaking communities.

### Dataset Details

1. **Domain-Specific Design**  
   Questions are crafted based on the _Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5)_, ensuring relevance to mental health diagnoses.

2. **High-Quality Annotations**  
   Each question includes multiple-choice options, the correct answer, and supporting paragraphs derived from trusted academic and medical sources, verified by mental health professionals.

3. **Alignment with Medical Standards**  
   Adhering to the _United States Medical Licensing Examination (USMLE)_ format, VMHQA ensures quality meets global medical standards.

4. **Advanced Question Types**  
   Features diverse and complex question types, challenging AI models to reason, comprehend, and apply knowledge effectively.

5. **Advancing Vietnamese NLP**  
   As the largest Vietnamese QA dataset in the mental health domain, VMHQA significantly contributes to the development of natural language processing (NLP) resources for a low-resource language.

---

## Dataset Overview

- Collected manually by human: 10,066
- Generated by GPT-4o: 2,200 records
- Generated by GPT-4o-mini: 5,000 records

---

## Dataset Structure

The dataset is provided in CSV format. Each row represents a single question-answer data point with the following columns:

- **Topic**: Main topic or category of the question.
- **Question**: The question text describing the scenario or problem.
- **Answer**: The correct answer option (1-based index corresponding to A, B, C, or D).
- **A, B, C, D**: The multiple-choice options for the question.
- **Paragraph 1, Paragraph 2, Paragraph 3**: Contextual paragraphs supporting the question and answers.
- **Fact 1, Fact 2, Fact 3**: Key supporting facts extracted from the paragraphs to justify the correct answer.
- **Question_Type**: The type of question (e.g., "Complex").
- **Reference**: Source link or reference for the information provided.
- **ID**: A unique identifier for each question-answer data point.
- **Sub_topic**: Specific subcategory of the topic.
- **Subject**: Broader subject area related to the topic.

This structured format ensures clarity and supports effective training and evaluation of AI models in the Vietnamese mental health domain.

---

## Dataset Creation

The VMHQA dataset was developed following these steps:

1. **Source Compilation**  
   Data was gathered from trusted academic books, medical websites, and hospital records.

2. **Annotation**  
   Psychology students crafted questions, answers, and supporting paragraphs following the USMLE format.

3. **Expert Verification**  
   The dataset was validated by **National Mental Hospital 2** to ensure accuracy and reliability.

4. **Data Refinement**

   - Removed duplicates and standardized formats.
   - Normalized topics and corrected inconsistencies.

5. **Final Output**  
   Produced 10,000 high-quality records, ensuring robustness and relevance for mental health applications.

---

## Dataset Demo

Sample examples of the dataset are available in the `Sample Data` folder.

- Each sub-dataset (origin, GPT-4o, GPT-4o-mini) contains 10 examples.
- Use these examples to familiarize yourself with the dataset structure and format.

---

## Dataset Access

1. **Sign the End User License Agreement (EULA) Form**  
   Download and sign the User Agreement form.
2. **Submit the Form**  
   Send the signed form to the dataset owner via the email address provided in the form.
3. **Receive Instructions**  
   Upon approval, you will be granted access to the full dataset on Hugging Face.

---

## Citation

When publishing research or presenting work that uses the VMHQA dataset, users must cite the dataset properly. The recommended citation in IEEE format is:

> T. A. H. Nguyen, Q.-D. Nguyen, H. M. Nguyen, A. H. Nguyen, and L. Nguyen,  
> “_VMHQA: A Vietnamese Multi-choice Dataset for Mental Health Domain Question Answering_,”  
> _EAI Endorsed Transactions on Scalable Information Systems_,  
> vol. 12, no. 4, Sep. 2025, doi: [10.4108/eetsis.7678](https://doi.org/10.4108/eetsis.7678)

Alternatively, you can cite it using the provided BibTeX entry:

```tex
@article{article,
  author  = {Nguyen, Loan and Nguyen, Tu and Nguyen, Quang-Dieu and Nguyen, Harius and Nguyen, Alfred},
  year    = {2025},
  month   = {09},
  pages   = {},
  title   = {VMHQA: A Vietnamese Multi-choice Dataset for Mental Health Domain Question Answering},
  volume  = {12},
  journal = {EAI Endorsed Transactions on Scalable Information Systems},
  doi     = {10.4108/eetsis.7678}
}
```

Failure to properly cite the dataset may result in revocation of access or denial of future dataset requests.

---

## Dataset Card Contact

For inquiries or support, please reach out to the dataset owner through the email provided in the EULA form.

---

## License

The dataset is distributed under specific terms outlined in the EULA form. Usage requires signing and adhering to the agreement's conditions.

---

Thank you for your interest in the **VMHQA Dataset**! This resource aims to support advancements in Vietnamese NLP and the mental health domain.