Bayanat Labs Arab-World Wikidata Slice
Facts about the places, organisations, people and works of the Arab world are spread across the whole of Wikidata. This dataset takes the 844,634 items tied to the 22 member states of the League of Arab States in the Wikidata JSON dump of 26 October 2026 and ships them as Parquet tables that join on the Wikidata id: one row per item with its class group, labels, descriptions and aliases in Arabic, Egyptian Arabic, Moroccan Arabic, English and French, and the states it is tied to and how; a long table of labels with a normalised form of the Arabic text for name matching; statements for 28 properties; titles on the Arabic, Egyptian Arabic, Moroccan Arabic, English and French Wikipedias, Arabic Wikiquote and Arabic Wikisource; and an Arabic Label Gap Index per state and class group, with lists of well-linked items that still lack an Arabic label. 536,400 of the items have an Arabic label, so 36.49% have none; by state the gap runs from 10.28% (Yemen) to 86.07% (Djibouti).
| Licence | CC0 1.0 (data, from Wikidata); Apache-2.0 (builder code); CC BY 4.0 (documentation in the GitHub repository); Wikimedia Commons files named in the image column keep their own licences |
|---|---|
| Version | v0.1.0 · Version v0.1.0, built from the Wikidata JSON dump of 26 October 2026; the builder is tagged on GitHub and not on PyPI yet. |
| Published | |
| Updated | |
| Who it is for | Teams that link Arabic text to entities, match Arabic and Latin names of places, organisations and people, build gazetteers or retrieval indexes for the Arab world, or measure where Wikidata still lacks Arabic labels. |
| Level | Intermediate |
Key numbers
Counted by script from the released files. Each key number is read from meta/summary.json in the Hugging Face dataset and pinned to its SHA-256; MANIFEST.json lists every file with its size and SHA-256.
Downloads and identifiers
| Where | What |
|---|---|
| Hugging Face | Dataset on Hugging Face |
| GitHub | Builder on GitHub |
| huggingface.co | Release numbers (meta/summary.json) |
| huggingface.co | Table checks and count comparisons (meta/validation.json) |
| GitHub | Method |
Get started
Load a config with the datasets library, or query the Parquet files directly with DuckDB. Install the builder from GitHub at the v0.1.0 tag to rebuild the slice from a Wikidata dump; the package is not on PyPI yet.
# Python
from datasets import load_dataset
entities = load_dataset("Bayanat-Labs/arab-world-wikidata", "entities", split="train")
coverage = load_dataset("Bayanat-Labs/arab-world-wikidata", "coverage", split="train")
saudi = entities.filter(lambda r: "Q851" in r["countries"])
# Builder
pip install "git+https://github.com/Bayanat-Labs/arab-world-wikidata@v0.1.0"
arab-world-wikidata stream --helpWhat is released
| Part | Licence | In this release |
|---|---|---|
| Items, labels, statements and sitelinks: configs entities, labels_long, triples and sitelinks (Parquet) | CC0 1.0 | Yes |
| Arabic Label Gap Index and gap lists: configs coverage, gap_todo and states (Parquet), with the coverage chart | CC0 1.0 | Yes |
| Build records: candidate list, dump checksums, selection and count queries, table checks and summary numbers | CC0 1.0 | Yes |
| Builder code: dump filter, build, checks and chart | Apache-2.0 | Yes |
| README, changelog and method in the GitHub repository | CC BY 4.0 | Yes |
| Images on Wikimedia Commons named in the image column (file names only; no image is included) | Each Commons file carries its own licence | No |
Attribution: Arab-World Wikidata Slice by Bayanat Labs. Data from Wikidata, CC0 1.0. https://huggingface.co/datasets/Bayanat-Labs/arab-world-wikidata
Intended uses
- Entity linking and named-entity work on Arabic text: link mentions to Wikidata ids with labels and aliases in Arabic, Egyptian Arabic, Moroccan Arabic, English and French.
- Name matching and gazetteers for places, organisations and people tied to the Arab League states, with the normalised Arabic form of each label.
- Measuring Arabic label coverage in Wikidata by state and class group, and finding well-linked items that still lack an Arabic label.
Out of scope
- Profiling individuals: the data about people is kept small and is not meant for profiling.
- A complete or verified register of the Arab world: Wikidata is uneven, and the slice follows its statements as they are.
- Settling borders or disputed areas: ties follow Wikidata statements, and Bayanat Labs takes no position on any boundary.
Rights at a glance
Training, display, redistribution and synthetic-voice use are separate permissions.
| Use | Status | Note |
|---|---|---|
| Training | Allowed | CC0 1.0, with no conditions. Only the names of Commons images are included, not the images. |
| Display | Allowed | CC0 1.0; "Data from Wikidata, CC0 1.0" is appreciated but not required. |
| Redistribution | Allowed | Data under CC0 1.0, builder code under Apache-2.0, documentation in the GitHub repository under CC BY 4.0. The Bayanat Labs name and logo are not licensed under CC0 1.0. |
| Synthetic voice | Not applicable | No audio. |
Limitations
- Wikidata is uneven: coverage differs by state and topic, and the gap index measures Wikidata, not the Arabic language. Large groups of similar items, such as mosques, weigh heavily on some totals; the coverage config splits the index by class group.
- Ties use truthy statements, so wrong or missing country, administrative-location or citizenship statements in Wikidata carry into the slice. Items in disputed areas can be tied to more than one state, or to a state other than the one a reader expects; the ties column shows which path tied each item to which state.
- The candidate list was made on 19 October 2026 and the values come from the dump of 26 October 2026: items tied to a state in between are not included, and candidates missing from the dump or no longer tied to a state are listed in the build records and left out.
- The present-day member states are the anchors, so an item tied only to a historical state is not selected. Labels in the multilingual default are not counted as Arabic, even in Arabic script.
- For people, only a small set of statements is released, without qualifiers, coordinates or IDs. Statements for dates of birth and death, religion, ethnic group, sexual orientation, medical conditions, political party and family relations are not released; birth dates are used only for the age filter. Labels and descriptions are copied from Wikidata as they are and can mention a year of birth, a life span, an occupation or a relative.
- Install the builder from GitHub for now; the package is not on PyPI yet.
Sources and licences
| Source | Licence | Redistributed here |
|---|---|---|
| Wikidata JSON dump of 26 October 2026 (wikidata-20261026-all.json.gz) | CC0 1.0 (Wikidata structured data) | Yes |
| QLever Wikidata endpoint (University of Freiburg): the candidate list, made on 19 October 2026 | Query results used to list candidate items; the list of Wikidata ids is in the build records, and the values come from the dump | Yes |
| Wikidata Query Service: two count checks, 19 October 2026 | Counts used to check the build; the queries and their results are in the build records | No |
The values come only from the Wikidata JSON dump of 26 October 2026, streamed once from an official Wikimedia mirror (Academic Computer Club, Umeå University) and not stored; the checksums of the bytes read match those Wikimedia publishes. Each item's ties to the 22 states were worked out again from the dump's own statements. Image file names point to Wikimedia Commons, where each file carries its own licence; no image is included. Source details
How to cite
Cite as:
Bayanat Labs (2026). Arab-World Wikidata Slice, version 0.1.0 (Wikidata JSON dump of 2026-10-26). Bayanat Labs. https://huggingface.co/datasets/Bayanat-Labs/arab-world-wikidata
No DOI. Cite the URL in the citation above and the date you read it.
@misc{bayanatlabs2026arabworldwikidata,
author = {{Bayanat Labs}},
title = {Arab-World Wikidata Slice, version 0.1.0 (Wikidata JSON dump of 2026-10-26)},
year = {2026},
month = oct,
howpublished = {\url{https://huggingface.co/datasets/Bayanat-Labs/arab-world-wikidata}}
}
Changelog
| Date | Version | Change |
|---|---|---|
| v0.1.0 | Slice built from the Wikidata JSON dump of 26 October 2026: seven Parquet configs, with the Arabic Label Gap Index per state and class group; 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