fouadmahmoud281/wikipedia-Egyptian-finetuning-Dataset

Egyptian Arabic Wikipedia Dataset

Dataset Description

Egyptian Arabic Wikipedia Dataset is a comprehensive collection of articles from the Egyptian Arabic Wikipedia (ويكيبيديا مصرى), processed and structured for natural language processing tasks. The dataset focuses on the Egyptian dialect of Arabic (اللهجة المصرية), which is spoken by approximately 100 million people and is one of the most widely understood Arabic dialects.

Dataset Details

• Size: 10,000 articles (subset of the full Egyptian Arabic Wikipedia)
• Source: Egyptian Arabic Wikipedia
• Language: Egyptian Arabic (ARZ)
• Created: April 2025
• License: CC BY-SA 4.0
• Format: JSON/JSONL

Dataset Structure

Each article in the dataset contains:

{
  "id": "unique_hash_id",
  "page_id": 12345,
  "title": "عنوان المقالة",
  "full_text": "النص الكامل للمقالة...",
  "sections": [
    {
      "id": "introduction",
      "level": 0,
      "title": "Introduction",
      "content": "محتوى المقدمة..."
    },
    {
      "id": "section_1",
      "level": 1,
      "title": "عنوان القسم",
      "content": "محتوى القسم..."
    }
  ],
  "categories": ["تصنيف1", "تصنيف2"],
  "metadata": {
    "page_id": 12345,
    "revid": 987654,
    "timestamp": "2025-04-02T14:30:25.123Z",
    "links_count": 87,
    "templates_count": 12,
    "images_count": 15,
    "has_infobox": true
  },
  "linguistic_features": {
    "char_count": 15240,
    "word_count": 2850,
    "sentence_count": 142,
    "avg_sentence_length": 20.07
  },
  "quality": "high|medium|low",
  "url": "https://arz.wikipedia.org/wiki/عنوان_المقالة",
  "training_examples": [
    {
      "task": "task_type",
      "instruction": "تعليمات بالمصري",
      "input": "المدخلات",
      "output": "المخرجات المتوقعة",
      "quality": "high|medium"
    }
  ]
}

Specialized Datasets

The repository also includes specialized datasets derived from the main collection:

1. Title Generation Dataset: For training models to generate appropriate titles for text content
2. Section Generation Dataset: For creating content for specific article sections
3. Question Answering Dataset: Facts-based Q&A pairs in Egyptian Arabic
4. Summarization Dataset: Text summarization examples at different levels of detail
5. Classification Dataset: Topic categorization tasks for Egyptian Arabic text
6. Combined Instructional Dataset: A balanced collection of examples across all task types

Data Collection Process

The data was collected using the MediaWiki API to systematically extract content from the Egyptian Arabic Wikipedia. The collection process:

1. Retrieved article IDs and metadata using the allpages API endpoint
2. Extracted full article content, sections, and categories
3. Processed the content to remove markup and non-relevant elements
4. Generated derived training examples for various NLP tasks
5. Assessed article quality based on length, structure, and metadata

Intended Uses

This dataset is designed for:

• Pre-training or fine-tuning language models on Egyptian Arabic
• Developing NLP applications specific to the Egyptian dialect
• Linguistic research on dialectal Arabic
• Building classification, generation, or question-answering systems
• Creating educational tools for Egyptian Arabic language learners

Task-Specific Applications

The dataset includes examples for multiple NLP tasks:

• Text generation (titles, sections, completions)
• Question answering (factual and conversational)
• Summarization (short and detailed summaries)
• Classification (topic categorization)
• Entity extraction (identifying key terms and concepts)
• Dialectal text generation (Egyptian Arabic specific formulations)

Dataset Statistics

Category	Count
Total Articles	~10,000
High Quality Articles	~3,000
Medium Quality Articles	~4,500
Low Quality Articles	~2,500
Articles with Images	~4,000
Articles with Infoboxes	~3,500
Long Articles (>2000 words)	~1,500
Medium Articles (500-2000 words)	~3,500
Short Articles (<500 words)	~5,000

Citation

@dataset{egyptian_arabic_wikipedia_dataset,
  author = {Fouad Mahmoud},
  title = {Egyptian Arabic Wikipedia Dataset},
  year = {2025},
  publisher = {Hugging Face},
  journal = {Hugging Face Datasets},
  howpublished = {\url{https://huggingface.co/datasets/fouadmahmoud281/wikipedia-Egyptian-finetuning-Dataset}}
}

Ethical Considerations

• Content Warning: As with any encyclopedic content, some articles may contain sensitive topics or controversial material.
• Bias: The dataset may reflect biases present in the original Wikipedia content and the Egyptian Arabic-speaking community.
• Quality Variation: Article quality varies considerably across the dataset, with some entries being more comprehensive than others.
• License Compliance: Usage should comply with the CC BY-SA 4.0 license, which requires attribution and sharing under the same terms.

Acknowledgements

This dataset was derived from the Egyptian Arabic Wikipedia, maintained by volunteer editors across the Arabic-speaking world. We acknowledge the Wikimedia Foundation and all contributors to the Egyptian Arabic Wikipedia for creating this valuable resource.

The dataset processing tools were developed using the MediaWiki API and open-source Python libraries including requests, BeautifulSoup, pandas, and tqdm.

Examples

Example 1: Article on Cairo (القاهرة)

{
  "id": "a1b2c3d4e5f6g7h8i9j0",
  "title": "القاهرة",
  "full_text": "القاهره هى عاصمة مصر و اكبر مدينه فى افريقيا و الشرق الاوسط...",
  "sections": [
    {
      "id": "introduction",
      "title": "Introduction",
      "content": "القاهره هى عاصمة مصر و اكبر مدينه فى افريقيا و الشرق الاوسط..."
    },
    {
      "id": "section_1",
      "title": "تاريخ",
      "content": "تأسست القاهره سنة 969 على يد جوهر الصقلى..."
    }
  ],
  "categories": ["عواصم عربية", "مدن مصرية", "محافظة القاهرة"],
  "quality": "high"
}

Example 2: Title Generation Task

{
  "task": "title_generation",
  "instruction": "اقرأ النص التالي ثم اقترح عنوان مناسب له:",
  "input": "هى عاصمة مصر و اكبر مدينه فى افريقيا و الشرق الاوسط. تأسست سنة 969...",
  "output": "القاهرة",
  "quality": "high"
}

Example 3: Question Answering Task

{
  "task": "factual_qa",
  "instruction": "أجب عن السؤال التالي بدقة وبناءً على المعلومات المتاحة:",
  "input": "ايه هو/مين هو القاهرة؟",
  "output": "القاهره هى عاصمة مصر و اكبر مدينه فى افريقيا و الشرق الاوسط...",
  "quality": "high"
}

How to Use

from datasets import load_dataset

# Load the main dataset
dataset = load_dataset("fouadmahmoud281/wikipedia-Egyptian-finetuning-Dataset")

# Access an article
article = dataset['train'][0]
print(f"Title: {article['title']}")
print(f"Introduction: {article['sections'][0]['content'][:200]}...")

# Load a task-specific dataset
qa_dataset = load_dataset("fouadmahmoud281/wikipedia-Egyptian-finetuning-Dataset", "question_answering")
print(f"Question: {qa_dataset['train'][0]['input']}")
print(f"Answer: {qa_dataset['train'][0]['output'][:100]}...")

Limitations

• The dataset represents only the content available in the Egyptian Arabic Wikipedia as of April 2025
• Article length and quality vary significantly across the dataset
• Some articles may contain formatting artifacts from the original wiki markup
• The dialect identification is based on the overall language of Egyptian Wikipedia, but individual articles may contain standard Arabic portions or other Arabic dialects