Bayanat Labs Arabic Wikipedia Monthly
Clean text from Arabic, Egyptian Arabic and Moroccan Arabic Wikipedia, built from the official Wikimedia dumps of 1 October 2026. Each record is one article at one revision, with a link to the article and to the exact revision, its Wikidata id, the text split into sections, infobox key-value pairs, tables, cited URLs, categories, topic labels, revision statistics and a rule-based flag that separates template-generated articles from individually written ones. Version v0.1.0 holds every Egyptian Arabic and Moroccan Arabic article with running text and a seeded random sample of Arabic Wikipedia; a separate labels config holds the identifiers and labels without text.
| Licence | CC BY-SA 4.0 (article text, from Wikipedia); CC BY 4.0 (labels config and documentation); Apache-2.0 (builder code) |
|---|---|
| Version | v0.1.0 · Version v0.1.0, built from the Wikimedia dumps of 1 October 2026; the builder is tagged on GitHub and not on PyPI yet. |
| Published | |
| Updated | |
| Who it is for | Teams that pretrain, adapt or evaluate models on Arabic, Egyptian Arabic or Moroccan Arabic text and need clean encyclopedic text with revision links, structure and a way to tell template-generated articles from individually written ones. |
| Level | Intermediate |
Key numbers
Counted by script from the released files. Each key number is read from stats.json in the Hugging Face dataset and pinned to its SHA-256; MANIFEST.json lists every file with its SHA-256, and PROVENANCE.jsonl records each dump file with its checksums.
Downloads and identifiers
| Where | What |
|---|---|
| Hugging Face | Dataset on Hugging Face |
| GitHub | Builder on GitHub |
| huggingface.co | Release statistics (stats.json) |
| Hugging Face | Sources and licences |
Get started
Load a config with the datasets library, or install the builder from GitHub at the v0.1.0 tag to rebuild a wiki from a Wikimedia dump; the package is not on PyPI yet.
# Python
from datasets import load_dataset
arz = load_dataset("Bayanat-Labs/arabic-wikipedia-monthly", "arz", split="train")
written = arz.filter(lambda r: r["generation"] == "human")
# Builder: Moroccan Arabic Wikipedia, dump of 1 October 2026
pip install "git+https://github.com/Bayanat-Labs/arabic-wikipedia-monthly@v0.1.0"
arabic-wikipedia-monthly fetch --wiki arywiki --date 20261001
arabic-wikipedia-monthly bots --out bot_groups.json
arabic-wikipedia-monthly build --wiki arywiki --date 20261001
arabic-wikipedia-monthly export --date 20261001 --wiki arywikiWhat is released
| Part | Licence | In this release |
|---|---|---|
| Article text with sections, infoboxes, tables, cited URLs, categories, topics and labels: configs ar_sample, arz and ary (Parquet) | CC BY-SA 4.0 | Yes |
| Labels config: page and revision ids, Wikidata ids, generation flag and rule, revision statistics, template-family and near-duplicate groups, topics; no article text | CC BY 4.0 | Yes |
| Builder code: fetch, build, export and the text-recovery check | Apache-2.0 | Yes |
| Dataset card, sources and licence notes, and documentation | CC BY 4.0 | Yes |
Attribution: Arabic Wikipedia Monthly by Bayanat Labs, text from Wikipedia contributors under CC BY-SA 4.0 (each record links to its article and revision). https://huggingface.co/datasets/Bayanat-Labs/arabic-wikipedia-monthly
Intended uses
- Pretraining or continued pretraining on encyclopedic text in Arabic, Egyptian Arabic and Moroccan Arabic, keeping or dropping template-generated articles with the generation flag.
- Building retrieval and question-answering corpora from sections, infoboxes, tables and Wikidata ids.
- Measuring how much of a Wikipedia edition was created in bulk from templates, with the rule behind each flag.
Out of scope
- A full copy of Arabic Wikipedia: this version holds a sample of it.
- Labels written by people: the generation flag and the topics come from rules and the category graph.
Rights at a glance
Training, display, redistribution and synthetic-voice use are separate permissions.
| Use | Status | Note |
|---|---|---|
| Training | Allowed | Text under CC BY-SA 4.0 and labels under CC BY 4.0, with attribution; each record links to its article and revision. |
| Display | Allowed | Text under CC BY-SA 4.0, with attribution and a link to the licence; text you share, adapted or not, carries CC BY-SA 4.0. |
| Redistribution | Allowed | Text under CC BY-SA 4.0 (share-alike), labels under CC BY 4.0, builder code under Apache-2.0. Wikipedia is a trademark of the Wikimedia Foundation; this dataset is not affiliated with or endorsed by it. |
| Synthetic voice | Not applicable | Text only; no audio. |
Limitations
- Text that a page shows only through templates (navigation boxes, glossaries, lists filled by templates) is not extracted, so recall is lower on template-built pages.
- The generation flag and the topics come from rules and the category graph; they have not been checked against labels written by people, and some topic labels will be wrong.
- The Arabic Wikipedia part is a sample: counts, template families and duplicate groups inside it are not those of the whole wiki.
- Wikipedia text reflects the coverage and viewpoints of its editors, and a revision can contain vandalism that was later reverted. Biographies of living people are included as published on Wikipedia; contributor names are not included.
- Install the builder from GitHub for now; the package is not on PyPI yet.
Sources and licences
| Source | Licence | Redistributed here |
|---|---|---|
| Arabic Wikipedia (arwiki), Wikimedia dumps of 1 October 2026 | CC BY-SA 4.0 (earlier contributions also CC BY-SA 3.0 and the GFDL); credit the contributors through the article history | Yes |
| Egyptian Arabic Wikipedia (arzwiki), Wikimedia dumps of 1 October 2026 | CC BY-SA 4.0 (earlier contributions also CC BY-SA 3.0 and the GFDL); credit the contributors through the article history | Yes |
| Moroccan Arabic Wikipedia (arywiki), Wikimedia dumps of 1 October 2026 | CC BY-SA 4.0 (earlier contributions also CC BY-SA 3.0 and the GFDL); credit the contributors through the article history | Yes |
| Wikidata item ids (wikidata_qid), from the wikibase_item page property in the dumps | Wikidata data is available under CC0 1.0 | Yes |
| MediaWiki Action API of each wiki: bot-group lists and rendered revisions for the text-recovery check | Used to compute labels and checks; the responses themselves are not part of the release | No |
Article text comes only from the Wikimedia dumps; each dump file's SHA-1 was checked against the value Wikimedia publishes and its SHA-256 recorded. Each record links to its article, whose history lists the authors, and to the exact revision used. The text was changed from the original: wikitext was converted to plain text, and markup, most templates, references and boilerplate sections were removed. Wikidata item ids (wikidata_qid) come from the wikibase_item page property in the dumps; Wikidata data is available under CC0 1.0. Wikipedia is a trademark of the Wikimedia Foundation; this dataset is not affiliated with or endorsed by it. Source details
How to cite
Cite as:
Bayanat Labs (2026). Arabic Wikipedia Monthly, version 0.1.0 (Wikimedia dumps of 1 October 2026). Bayanat Labs. https://huggingface.co/datasets/Bayanat-Labs/arabic-wikipedia-monthly
No DOI. Cite the URL in the citation above and the date you read it.
@misc{bayanatlabs2026arabicwikipediamonthly,
author = {{Bayanat Labs}},
title = {Arabic Wikipedia Monthly, version 0.1.0 (Wikimedia dumps of 1 October 2026)},
year = {2026},
month = oct,
howpublished = {\url{https://huggingface.co/datasets/Bayanat-Labs/arabic-wikipedia-monthly}}
}
Changelog
| Date | Version | Change |
|---|---|---|
| v0.1.0 | Release built from the Wikimedia dumps of 1 October 2026: all Egyptian Arabic and Moroccan Arabic Wikipedia articles with running text, a seeded random sample of Arabic Wikipedia and a labels config; builder on GitHub. |
Corrections and removal requests
If something in this release should not be there, or is wrong, email hello@bayanatlabs.com. We handle removal requests within 30 days, and platform requests within 24 hours. How removal requests work
Related product: Bayanat Knowledge