Palestinian Cultural Knowledge Corpus
v0.2.0 — supersedes the earlier data/wikipedia_ar/ v0.1.0 partial upload
(484 Arabic Wikipedia documents only). This release expands to the full 5-source
corpus below and moves the data to data/full_corpus/.
A multi-source Arabic/English text corpus about Palestinian history, culture, and heritage, built for the Palestinian Cultural Knowledge Platform — a RAG + knowledge-graph research project. 882 documents, ~890K words, collected and processed as of 2026-08.
Dataset composition
| Source | Documents | Notes |
|---|---|---|
| Wikipedia (Arabic) | 581 | Full article text |
| WAFA (وكالة وفا, Palestinian news agency) | 106 | Full article text |
| Wikipedia (English) | 97 | Full article text |
| GDELT-indexed news | 73 | Full article text, fetched from the outlet GDELT indexed |
| Semantic Scholar | 25 | Title + abstract only (see rights note below) |
Language split: 712 Arabic (ar-MSA), 131 English, 39 unknown/mixed.
Each record carries a source_id field so the dataset can be filtered to a subset —
e.g. source_id in {"wikipedia-ar", "wikipedia-en", "semantic-scholar"} for the subset
with the clearest redistribution rights (see below).
Licensing & rights
Per-source assessment (full reasoning in the source project's
docs/licensing_checklist.md):
| Source | Status | Basis |
|---|---|---|
| Wikipedia (ar/en) | Clear | CC BY-SA 4.0 — reuse/redistribution with attribution and share-alike |
| Semantic Scholar (title+abstract) | Clear | Permitted under the S2AG API's terms of use for research/product use |
| WAFA | Needs review | robots.txt permits crawling, but that governs indexing, not redistribution; WAFA's articles are copyrighted news content and no explicit reuse license was found |
| GDELT | Needs review | GDELT's own event metadata is freely reusable, but the article text here was fetched from the outlet GDELT pointed to — a separate, per-outlet rights question GDELT's license doesn't resolve |
This release includes all five sources, including the two marked "needs review."
That's a deliberate choice by this dataset's publisher, made with the rights gap above
known and disclosed — not an oversight. If you need a subset with confirmed
redistribution rights only, filter to source_id in {"wikipedia-ar", "wikipedia-en", "semantic-scholar"} (703 of the 882 documents).
Collection & processing
- Wikipedia: official MediaWiki API.
- WAFA: direct site collection (
wafa.ps). - GDELT: GDELT Project's event index, article text fetched from each indexed outlet.
- Semantic Scholar: S2AG public API, metadata fields only (no PDF text).
Documents passed through language detection, deduplication (MinHash), and a
rule/LLM-based quality and relevance gate before inclusion — see the source repo's
src/ingestion/ for the full pipeline. category, topic_id/topic_label, and
entities fields come from later analysis passes (zero-shot LLM classification and
BERTopic respectively) and are included as-is, not guaranteed error-free — see the
source repo's ROADMAP.md Track F for known accuracy caveats.
Files
train.parquet— the full 882-row table (recommended fordatasets.load_dataset).train.jsonl— the same data, JSON Lines.dataset_summary.json— column list and per-source/per-language counts.
Intended use
Research and educational use: RAG systems, NLP research on Arabic/Palestinian content, topic/bias analysis. Given the "needs review" status of two sources above, downstream redistribution of this dataset (as opposed to using it directly) should independently confirm rights for the WAFA/GDELT-sourced rows, or filter them out first.
- Downloads last month
- 7