☪ Chinese-Muslim-Travel: Native Chinese Muslim Travel RAG Corpus

Dataset Description

Chinese-Muslim-Travel is a curated RAG corpus containing 347 native Chinese articles documenting Muslim travel, halal food, mosque architecture, and Muslim community life across 20+ countries. Every article is original native Chinese text — no machine translation, no synthetic generation. All 23,192 embedded image references are preserved.

Key Properties

Property	Value
Total Records	347
Language	Chinese (Simplified, zh-CN)
Format	Parquet (auto Dataset Viewer enabled)
Splits	train (277), validation (34), test (36)
Avg. Article Length	6,911 characters
Total Characters	2,398,343
Total Image References	23,192
Articles with Images	345 / 347
License	MIT
Version	v2.0.0
Release Date	2026-06-13

---

Embeddable Dataset Viewer

Embed in any webpage:

<iframe
  src="https://huggingface.co/datasets/qurancn/Chinese-Muslim-Travel/embed/viewer/default/train"
  frameborder="0"
  width="100%"
  height="560px"
></iframe>

Direct viewer links:

• 🔍 Browse All (train)
• 🕌 Mosques & Architecture
• 🍖 Halal Food
• ✈️ Travel

---

Topic Distribution

Category	Count
Culture & Life	81
Mosques & Architecture	79
Travel & Exploration	69
History & Heritage	67
Halal Food & Cuisine	38
Muslim Communities	13

---

Data Schema

Column	Type	Description	Queryable
id	string	Article ID from source platform	✅
title	string	Original Chinese article title	✅ Full-text
text	string	Complete article body with Markdown images	✅ Full-text
text_preview	string	First 300 chars (Viewer-friendly preview)	✅
source_url	string	Canonical URL on salaamalykum.com	✅
content_hash	string	SHA-256 integrity hash	✅
language	string	BCP-47 language tag (zh-CN)	✅ Filter
topic_category	string	One of 6 categories	✅ Filter
char_count	int32	Character count of full article body	✅ Sort
word_count	int32	Word/token count	✅ Sort
image_count	int32	Number of embedded image references	✅ Sort
has_images	bool	Whether article contains images	✅ Filter
source_platform	string	Source website identifier	✅ Filter

Note on images: All image URLs are preserved inline as Markdown ![image](url) within the text field. Images are hosted externally (archive.org, catbox.moe, etc.) and render when viewing the raw Markdown.

SQL Console Queries

Run these directly in Data Studio:

-- Count articles by topic
SELECT topic_category, COUNT(*) as count, AVG(char_count) as avg_length
FROM train GROUP BY topic_category ORDER BY count DESC;

-- Top 10 longest articles
SELECT title, char_count, image_count, source_url
FROM train ORDER BY char_count DESC LIMIT 10;

-- Articles with most images
SELECT title, image_count, char_count
FROM train WHERE has_images = true ORDER BY image_count DESC LIMIT 10;

-- Search for specific mosque articles
SELECT title, text_preview, source_url
FROM train WHERE title LIKE '%清真寺%';

---

Data Samples

Example: 2016年的牛街筛海坟与文化陈列室

Field	Value
id	1898
title	2016年的牛街筛海坟与文化陈列室
text_preview	网络转帖 摘要：这篇文章围绕“2016年的牛街筛海坟与文化陈列室”展开，适合关注穆斯林生活、清真信息、伊斯兰文化和中文网络转帖内容的读者阅读。 2016年的牛街筛海坟与文化陈列室 2016年一次去牛街礼拜寺的时候拍了下里面的筛海坟和文物陈列室，当时手机拍照效果一般，本来...
source_url	https://salaamalykum.com/cn/article/1898
topic_category	Culture & Life
char_count	5105
image_count	39

Example: 2017年的江苏淮安

Field	Value
id	1699
title	2017年的江苏淮安
text_preview	网络转帖 摘要：这篇文章围绕“2017年的江苏淮安”展开，适合关注穆斯林生活、清真信息、伊斯兰文化和中文网络转帖内容的读者阅读。 2017年的江苏淮安 因文章被下架，现修改后重新发出。 2017年3月11日，我去江苏淮安逛吃，去了清江浦、河下和王家营三处，下面就分别介...
source_url	https://salaamalykum.com/cn/article/1699
topic_category	Culture & Life
char_count	9862
image_count	97

Example: 2018年的山海关

Field	Value
id	1842
title	2018年的山海关
text_preview	网络转帖 摘要：这篇文章围绕“2018年的山海关”展开，适合关注穆斯林生活、清真信息、伊斯兰文化和中文网络转帖内容的读者阅读。 2018年的山海关 2018年冬日的阳光非常明媚，我坐早班高铁从北京南站两个小时就到了山海关。山海关人说话尾音和葫芦岛、锦州人一样说话往上扬，...
source_url	https://salaamalykum.com/cn/article/1842
topic_category	Culture & Life
char_count	5813
image_count	85

---

Data Collection & Processing Pipeline

Source

All 347 articles originate from salaamalykum.com/cn/topic/穆斯林旅行, a community-driven platform documenting Chinese Muslim travel and life.

Extraction Method

Articles were extracted directly from the source platform's MySQL database (WeCenter 3.1.9, aws_article table joined via aws_topic_relation with topic_id=838), bypassing a frontend pagination bug that limited web scraping to only 23 articles. Database extraction recovered all 347 articles with complete content.

Processing Steps

1. SQL Dump Parsing: Custom Python state-machine parser extracts tuples from INSERT INTO aws_article VALUES(...) statements in a 89MB MySQL dump
2. Topic Relation Mapping: aws_topic_relation table maps article IDs to topic_id=838 (穆斯林旅行), yielding exactly 347 matches
3. BBCode → Markdown: [img]url[/img] → ![image](url), [b]text[/b] → **text**, [url=...]text[/url] → [text](url)
4. HTML Entity Cleanup: html.unescape() + <br> → \n normalization
5. SHA-256 Integrity Hash: Computed on final cleaned body text for each article
6. Topic Classification: Rule-based keyword classifier assigns one of 6 categories based on title keywords
7. Parquet Serialization: pyarrow with row_group_size=50, write_page_index=True for optimal HF Viewer performance
8. Split Generation: Deterministic shuffle (seed=42) then 80/10/10 split into train/validation/test

Known Limitations & Failure Boundaries

• No machine translation: All text is native Chinese — intentional design choice
• Image hosting: ~23,192 image URLs reference external hosts (archive.org, catbox.moe); images are not embedded as binary in the dataset
• Author attribution: Individual author names not preserved (WeCenter DB schema limitation)
• Temporal coverage: Articles span approximately 2020–2026
• Frontend bug: The source platform's web frontend only exposes 23/347 articles via pagination; full recovery required direct database access
• Encoding edge cases: A small number of articles may contain residual escape sequences from MySQL dump parsing

---

Usage

Python (Hugging Face Datasets)

from datasets import load_dataset

ds = load_dataset("qurancn/Chinese-Muslim-Travel")
print(ds)
# DatasetDict({
#     train: Dataset({features: [...], num_rows: 277})
#     validation: Dataset({features: [...], num_rows: 34})
#     test: Dataset({features: [...], num_rows: 36})
# })

# Access full article text with all images
article = ds["train"][0]
print(article["title"])  # Chinese title
print(article["text"])   # Full body with ![image](...) references
print(article["image_count"])  # Number of images in this article

LangChain RAG Pipeline

from langchain.vectorstores import FAISS
from langchain.embeddings import HuggingFaceEmbeddings
from datasets import load_dataset

ds = load_dataset("qurancn/Chinese-Muslim-Travel", split="train")
texts = [f"{row['title']}\n{row['text']}" for row in ds]

embeddings = HuggingFaceEmbeddings(model_name="shibing624/text2vec-base-chinese")
vectorstore = FAISS.from_texts(texts, embeddings)

Filter by Topic

mosques = ds["train"].filter(lambda x: x["topic_category"] == "Mosques & Architecture")
food = ds["train"].filter(lambda x: x["topic_category"] == "Halal Food & Cuisine")

---

Version History

Version	Date	Changes
v2.0.0	2026-06-13	Full rebuild: 347 articles in Parquet, 3 splits, complete text + images, 13-column schema
v1.0.0	2026-06-13	Initial upload (legacy corpus.parquet + rag_chunks.parquet)

---

Citation

@dataset{chinese_muslim_travel_2026,
  title={Chinese-Muslim-Travel: A Native Chinese RAG Corpus of 347 Articles on Muslim Travel and Halal Tourism},
  author={Salaamalykum Project},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/qurancn/Chinese-Muslim-Travel}
}

---

License

MIT License

Contact

• Email: bropeace@protonmail.com
• GitHub: https://github.com/salaamalykum
• Website: https://salaamalykum.com