MuslimNews: Global Islamic & International News RAG Corpus (v1.0.0) 🌍🕌

MuslimNews is a rigorously curated, zero-noise Chinese corpus containing 173 verified international news articles and 1092 semantic RAG chunks covering global Muslim communities, Middle East & South Asia public policy, religious heritage protection, Halal economy, and cross-cultural dialogues.

Synchronized directly from https://salaamalykum.com/cn/topic/国际新闻, this dataset follows a strict Dual-Track Publishing Architecture:

• Human-Readable Portal & Markdown Archive: Hosted on GitHub (salaamalykum/muslimnews) and GitHub Pages (https://salaamalykum.github.io/muslimnews/).
• Machine-Trainable Views: Hosted here on Hugging Face (qurancn/muslimnews) in page-indexed Apache Parquet (data/*.parquet) for instant Dataset Viewer / DuckDB SQL querying, alongside derived RAG Chunks, Alpaca SFT, and ShareGPT JSONL views in sft/.

Human Readers: Looking for the full text with all images rendered natively? Navigate to Files and versions -> content or visit our GitHub Pages Portal!

---

Dataset Views & Format Integrity

To uphold strict data quality standards, we separate the Primary Document RAG View from Derived SFT Views:

1. Primary Document RAG View (default Parquet config in data/*.parquet & sft/muslimnews_dataset.jsonl):
   • Contains the complete, unmanipulated 173 news articles with SHA-256 content hashes, original URLs, canonical URLs, authors, ISO-8601 timestamps, and topic tags.
2. Semantic Sliding-Window RAG Chunks (sft/muslimnews_rag_chunks.jsonl):
   • 1092 overlapping semantic chunks (chunk_id, article_id, title, text, original_url, content_hash) engineered for hybrid BM25 + Dense Vector Retrieval and cross-corpus alignment with quran_rag_dataset.
3. Instruction-Tuning Views (sft/muslimnews_alpaca.jsonl & sft/muslimnews_sharegpt.jsonl):
   • Explicitly structured single-turn (instruction / input / output) and multi-turn (conversations array) records derived from each article's editorial lead (【导读与核心背景】), summary (【核心内容概要】), and full report body.

Data Schema (default Parquet Config)

Field	Type	Description
id	string	Original WeCenter article ID from salaamalykum.com.
slug	string	URL-safe deterministic identifier ({id}_{sanitized_title}).
title	string	Headline of the international news article.
summary	string	Clean 220-character editorial lead summary.
text	string	Full zero-noise GitHub Flavored Markdown body.
original_url	string	Source URL (https://salaamalykum.com/cn/article/{id}).
canonical_url	string	GitHub Pages canonical URL.
author	string	Original contributor username.
pub_date	string	ISO-8601 publication timestamp (UTC).
updated_at	string	Synchronized batch release timestamp (2026-10-30T07:50:00Z).
language	string	BCP-47 language tag (zh-CN).
tags	string	Comma-separated topic tags mapped from aws_topic.
content_hash	string	SHA-256 cryptographic digest of text.
word_count	int64	Character length of text.
preview_excerpt	string	150-character snippet for rapid UI & Embedding Atlas rendering.

---

Quickstart Usage

from datasets import load_dataset

# Load the primary RAG document corpus
ds = load_dataset("qurancn/muslimnews", split="train")
print(ds[0]["title"], ds[0]["original_url"])

Citation

@dataset{muslimnews_rag_2026,
  author       = {Salaamalykum Open Data Hub},
  title        = {MuslimNews: A Curated RAG Corpus of 173 International News Articles on the Global Muslim World},
  month        = oct,
  year         = 2026,
  publisher    = {Hugging Face & GitHub},
  version      = {v1.0.0},
  url          = {https://github.com/salaamalykum/muslimnews}
}