moujar/MentalHealth-Darija

Dataset Card for Mental Health Darija

Dataset Summary

Mental Health Darija is a multilingual (English and Moroccan Darija) dataset for mental health text classification. Each example contains parallel text: an English sentence (text_en) and its Darija (Moroccan Arabic) translation (text), with a single label indicating the mental health category. The dataset supports classification into seven categories: Anxiety, Bipolar, Depression, Normal, Personality disorder, Stress, and Suicidal. It is intended for research and applications in mental health detection and support in under-resourced languages.

Supported Tasks and Leaderboards

• Text classification (multi-class): predict the mental health category from either the English or Darija text (or both).
• Suitable for benchmarking emotion/mental-health classification and for training or fine-tuning models on Darija.

Languages

• English (text_en): source or translated statements.
• Moroccan Darija (text): Moroccan Arabic dialect; primary target language for the dataset.

Dataset Structure

Data Instances

An example looks as follows.

{
  "text_en": "I've been so restless lately.",
  "text": "كنحس براسي مقلق هاد الأيام",
  "label": "Anxiety"
}

Data Fields

Field	Type	Description
text_en	string	English text (statement or translation).
text	string	Moroccan Darija text (statement or translation).
label	string	Mental health category (one of 7 classes).

Label set: Anxiety, Bipolar, Depression, Normal, Personality disorder, Stress, Suicidal.

Text length (characters): Darija text — min 2, max 1571, mean ~275; English text_en — min 2, max 32759, mean ~576.

Data Splits

The dataset is provided as a single split (no predefined train/validation/test split). Total: 51,093 examples.

label	count
Anxiety (0)	3,623
Normal (1)	16,040
Depression (2)	15,094
Suicidal (3)	10,644
Stress (4)	2,296
Bipolar (5)	2,501
Personality disorder (6)	895
Total	51,093

Recommended: create your own train/validation/test split (e.g. 80/10/10) with stratification on label for balanced evaluation.

Dataset Creation

Curation Rationale

The dataset was created to support mental health detection and NLP research in Moroccan Darija, a low-resource dialect, by providing parallel English–Darija text with mental health labels.

Source Data

Initial Data Collection and Normalization

More Information Needed

Who are the source language producers?

More Information Needed

Annotations

Annotation process

Labels correspond to mental health categories (Anxiety, Bipolar, Depression, Normal, Personality disorder, Stress, Suicidal). More Information Needed on annotation protocol and guidelines.

Who are the annotators?

More Information Needed

Personal and Sensitive Information

The dataset contains mental health–related text. It may include sensitive or distressing content. Use in accordance with ethical guidelines and applicable data protection regulations.

Considerations for Using the Data

Social Impact of Dataset

This dataset can support tools for mental health awareness and triage in Darija-speaking communities. Models trained on it should be used as research or decision-support aids only, not as substitutes for professional care.

Discussion of Biases

Class distribution is imbalanced (e.g. Depression and Normal have the most examples; Personality disorder has the fewest). Consider stratified sampling, class weights, or resampling when training and evaluating models.

Other Known Limitations

• No official train/validation/test split; users should define their own.
• Text length in text_en is highly variable (including some very long entries); preprocessing or truncation may be needed for fixed-length models.

Additional Information

Dataset Curators

Abderrahmane Moujar and Otmane El Bourki. Dataset repository: moujar,oelbourki on Hugging Face.

Licensing Information

This dataset is licensed under the MIT License. Use for educational and research purposes is encouraged; ensure compliance with the license and any applicable ethics and privacy requirements.

Citation Information

If you use this dataset, please cite the dataset repository:

@misc{mental-health-darija,
  author = {Abderrahmane Moujar and Otmane El Bourki},
  title = {Mental Health Darija},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/moujar/MentalHealth-Darija},
}

Contributions

Contributions to the dataset card or dataset are welcome. Please open an issue or pull request on the dataset repository.