ArabicNLPWorld/arabic-wikipedia-wikibooks-corpus

Arabic Wiki Corpus - Parquet Format

Dataset Description

This dataset contains CLEANED Arabic text from Wikipedia and Wikibooks in Parquet format (efficient, fast, columnar storage).

Key Features

• ✅ Parquet format - faster loading, smaller size, columnar storage
• ✅ Fully cleaned - no markup, no HTML, no references
• ✅ Arabic normalized - alef, ya, ta marbuta normalized
• ✅ Ready for ML/NLP/LLM - use directly without preprocessing

Statistics

Metric	Value
Total Articles	1,210,037
Wikipedia Articles	1,208,932
Wikibooks Articles	1,105
Unique Categories	642,128
Average Original Length	2,004 chars
Average Cleaned Length	1,836 chars
Average Compression	7.7%
Total Size (Original)	2.26 GB
Total Size (Cleaned)	2.07 GB
Format	Parquet (columnar)
Chunks	25 files

Loading the Dataset

Using Hugging Face datasets (Recommended)

from datasets import load_dataset

# Load directly from Parquet files
dataset = load_dataset("ArabicNLPWorld/arabic-wikipedia-wikibooks-corpus", split="train")

# Access cleaned text directly
for record in dataset:
    clean_text = record["text"]  # Already cleaned!
    print(record["title"])
    print(clean_text[:500])

Using PyArrow (Direct Parquet)

import pyarrow.parquet as pq
from huggingface_hub import hf_hub_download

# Download and read Parquet file
parquet_path = hf_hub_download(
    repo_id="ArabicNLPWorld/arabic-wikipedia-wikibooks-corpus",
    filename="data/chunk_0000.parquet",
    repo_type="dataset"
)
table = pq.read_table(parquet_path)
df = table.to_pandas()

Data Fields

Field	Type	Description
id	string	Unique MD5 hash
title	string	Article title
text	string	CLEANED text (ready for ML)
source	string	wikipedia / wikibooks
url	string	Original URL
categories	list	Category tags
meta_original_length	int	Length before cleaning
meta_cleaned_length	int	Length after cleaning
meta_compression_ratio	float	% of text removed

Example Record

{
  "id": "f2e322039571bb760fb615daa77f6adb",
  "title": "ماء",
  "text": "الماء مادهٌ شفافهٌ عديمه اللون والرائحه، وهو المكوّن الاساسي للجداول والبحيرات والبحار والمحيطات وكذلك للسوائل في جميع الكائنات الحيّه، وهو اكثر المركّبات الكيميائيّه انتشاراً علي سطح الارض. يتالّف جزيء الماء من ذرّه اكسجين مركزيه ترتبط بها ذرّتا هيدروجين علي طرفيها برابطه تساهميّه بحيث تكون صيغته الكيميائيه H2O. عند الظروف القياسيه من الضغط ودرجه الحراره يكون الماء سائلاً؛ امّا الحاله الصلبه فتتشكّل عند نقطه التجمّد، وتدعي بالجليد؛ امّا الحاله الغازيه فتتشكّل عند نقطه الغليان، وتسمّي بخار الماء",
  "source": "wikipedia",
  "url": "https://arwiki.org/wiki/ماء",
  "categories": [
    "مشروبات",
    "ماء",
    "مركبات الهيدروجين",
    "مركبات الأكسجين",
    "مذيبات لاعضوية"
  ],
  "meta_original_length": 61222,
  "meta_cleaned_length": 52814,
  "meta_compression_ratio": 13.73
}

Why Parquet?

Feature	JSONL	Parquet
File size	Larger	Smaller (compressed)
Load speed	Slower	Faster (columnar)
Memory usage	High	Low
Filtering	Full scan	Column pruning
Type safety	No	Yes (schema)

🤝 Citation

@misc{arabic_clean_parquet_ArabicNLPWorld,
  author = {Arabic NLP World Community},
  title = {Arabic Clean Corpus - Parquet Format},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/ArabicNLPWorld/arabic-wikipedia-wikibooks-corpus}
}

License

CC BY-SA 4.0

---

Generated with ❤️ for the Tajik NLP community