IbrahimAmin/egyptian-arabic-hate-speech

🇪🇬 Egyptian-Arabic Hate Speech Dataset 🗣️🚫

Author: IbrahimAmin, Mostafa Abbas, Rany Hatem, Andrew Ihab, Mohamed Waleed Fahkr
License: MIT
Paper: Fine-tuning Arabic Pre-Trained Transformer Models for Egyptian-Arabic Dialect Offensive Language and Hate Speech Detection and Classification
Languages: Arabic (Egyptian Dialect)

---

📋 Dataset Summary

This dataset consists of 8,169 Egyptian-Arabic text samples manually labeled for offensive language and hate speech classification. The samples were collected from:

• 🌐 Social media (Facebook, Twitter, etc.)
• 📚 Public datasets
• 📝 A custom Google survey

Each sample is annotated into one of five categories:

• 🟢 Neutral
• ⚠️ Offensive
• 🚺 Sexism
• ✝️ Religious Discrimination
• 🧬 Racism

---

📂 Dataset Structure

🧾 Data Fields

Each record contains:

• text (📝 string): The Egyptian-Arabic text sample
• label (📝 string): One of the following:
  • 🟢 Neutral
  • ⚠️ Offensive
  • 🚺 Sexism
  • ✝️ Religious Discrimination
  • 🧬 Racism

---

🔀 Dataset Splits

Split	📊 # of Samples
Train	6,535
Test	1,634
Total	8,169

🟰 Balanced class distribution in both sets.

---

🏗️ Dataset Creation

📥 Collection Sources

• Existing Arabic sentiment & hate speech datasets
• Manually curated Egyptian social media posts
• Custom 📝 Google survey with 250+ responses

🛠️ Preprocessing

• Normalize Arabic characters
• Remove tashkeel, tatweel, English/Hindi chars
• Strip special characters and stopwords

📈 Data Augmentation

Used for underrepresented classes (🚺, ✝️, 🧬) to ensure balance.

---

🧠 Intended Uses

• 💬 Hate speech detection in Arabic social media
• 🧪 Benchmarking Arabic transformer models
• 🛡️ Moderation tools and NLP fairness research in MENA

---

🔧 Usage

from datasets import load_dataset

dataset = load_dataset("IbrahimAmin/egyptian-arabic-hate-speech")

# Access a sample
sample = dataset["train"][3453]
print(sample["text"])
print(sample["label"])

---

⚠️ Limitations

• 📍 The dataset focuses on Egyptian dialect and may not generalize to other Arabic dialects.
• 🧑‍⚖️ Despite careful manual annotation, biases may exist in labeling due to subjectivity in defining hate/offensive content.

---

⚠️ Disclaimer

This dataset is intended for research and content moderation purposes and is not meant to offend, harm, or promote discrimination against any individual or group. It is important to use this dataset responsibly and consider the context in which it is applied. Any offensive content from this dataset should be treated with caution and handled appropriately.

---

📝 Citation

Please cite the following publication if you use this dataset:

@INPROCEEDINGS{10009167,
  author={Ahmed, Ibrahim and Abbas, Mostafa and Hatem, Rany and Ihab, Andrew and Fahkr, Mohamed Waleed},
  booktitle={2022 20th International Conference on Language Engineering (ESOLEC)}, 
  title={Fine-tuning Arabic Pre-Trained Transformer Models for Egyptian-Arabic Dialect Offensive Language and Hate Speech Detection and Classification}, 
  year={2022},
  volume={20},
  pages={170-174},
  doi={10.1109/ESOLEC54569.2022.10009167}
}

---

⚖️ License

This dataset is distributed under the MIT License. Use freely with attribution.

---

🙏 Acknowledgements

This dataset and any accompanying models were developed as part of a 🎓 Graduation Project at the Faculty of Engineering, AASTMT — Computer Engineering Program.