Houdna-khilouf/Dz-Emotion

---
language:
- ar
pretty_name: Dz-Emotion
license: cc-by-4.0

task_categories:
- text-classification

task_ids:
- multi-class-classification

multilinguality: monolingual

size_categories:
- 1K<n<10K

tags:
- arabic
- algerian-arabic
- emotion-classification
- nlp
---

# Dz-Emotion: Algerian Dialect Dataset for Emotion Detection

## 📌 Dataset Description

**Dz-Emotion** is the first large-scale, manually annotated dataset for **emotion detection in Algerian Arabic dialect (Darija)**. The dataset consists of **6,000 social media comments**, collected from YouTube, Facebook, and Instagram, and labeled according to **Ekman’s six basic emotions**:(Anger, Sadness, Fear, Disgust, Happiness, Surprise)

The dataset is designed to support research in **Natural Language Processing (NLP)** for low-resource dialects, especially Algerian Arabic.

---

## 📄 Paper

For more information, please visit our paper:

👉 https://ieeexplore.ieee.org/document/11472633

---

## 🤖 Related Model

This dataset was used to train:

- Dz-EmoBERT
https://huggingface.co/Houdna-khilouf/Dz-EmoBERT

**Dz-EmoBERT** is a fine-tuned transformer model for emotion detection in Algerian dialect text, achieving **94.08% accuracy** on this dataset.

---

## 📊 Dataset Structure

The dataset is provided as a CSV file with the following columns:

| Column | Description |
|--------|------------|
| ID | Unique identifier for each comment |
| Text | The comment text (Algerian dialect) |
| Label | Emotion label |
| Source | Platform source (YouTube, Facebook, Instagram) |

---

## 📈 Dataset Statistics

- Total samples: **6,000**
- Classes: **6 emotions**
- Samples per class: **1,000 (balanced)**

### Emotion Distribution:
- Anger: 1000
- Sadness: 1000
- Fear: 1000
- Disgust: 1000
- Happiness: 1000
- Surprise: 1000

### Data Sources:
- YouTube: 53%
- Instagram: 29%
- Facebook: 18%

### Train/Test Split:
- Train: 80% (4,800 samples)
- Validation: 20% (1,200 samples)

---

## 🚀 Baseline Results

The dataset was used to fine-tune several models:

| Model | Accuracy |
|-------------|---------|
| ARBERT | 86.00% |
| MARBERT | 91.67% |
| Dz-EmoBERT | **94.08%** |

---

## ⚠️ Limitations

- Data collected from social media may include noise and bias
- Focused only on **six emotions (Ekman model)**
- Limited to **Algerian dialect**

---

## 📩 Contact

For any questions or collaboration opportunities:

[e-mail]

---

## 📚 Citation

If you use this dataset, please cite:

```bibtex
@inproceedings{khilouf2025dzemotion,
title={Dz-Emotion: An Algerian Dialect Dataset for Text-Based Emotion Detection},
author={Khilouf, Houdna and Ziani, Amel and Malek, Nada Ahmed and Schwab, Didier and Yakoubi, Mohamed Amine},
booktitle={2025 International Conference on Recent Advances in Mathematics and Informatics (ICRAMI)},
pages={1--6},
year={2025},
address={Sousse, Tunisia},
doi={10.1109/ICRAMI64946.2025.11472633}
}