BAYANATبيانات Dataset & documentation
Open Arabic data · CC0

Arabic entities.
Names, connected.

An Arabic reference atlas for search and entity resolution. Explore source names, inspect ambiguity, and follow every entity back to Wikidata.

أسماء مختلفة، كيانات مترابطة.
اكتشف الاسم، وافهم الالتباس.

Get the complete dataset How the data works
—
entities in the complete dataset
—
source name-to-entity rows
—
ambiguous normalized keys
—
source-attested alias queries

These are full-release counts. The interactive explorer below searches a small embedded sample. Source names are not newly human-verified annotations.

Interactive sample explorer · مستكشف عينة

Sample only. No online corpus search or model inference. Every result comes from data embedded in this page.نتائج من العينة فقط، وليست بحثاً في كامل البيانات.
Focus search /
Explore a source name
Real source aliases
Matches within this sample
Normalized keys group source forms. A “complete ambiguity group” includes every candidate recorded for that key in this dataset; other groups show sample candidates only.
Built for inspection

A name is a candidate.
Context makes the link.

This explorer makes the structure visible. Use the complete Parquet release for training, indexing and reproducible evaluation.

01

Source names, preserved with display cleanup

Arabic labels and aliases come from Wikidata. Display cleanup and lossy search keys are separate. Names link to QIDs; English labels and descriptions appear only where the source supplies them.

02

Shared spellings, explicit candidates

Different entities can share one normalized name. Complete sampled ambiguity groups retain all of their released candidate QIDs. Candidate lists are possibilities, not contextual identity decisions.

03

Transparent sample, reusable full release

The embedded sample combines selected familiar labels and deterministic hash samples. It does not represent topic frequencies or full-dataset coverage. The dataset card documents filtering, splits, silver queries and limitations.