qurancn/Halal-Food-In-China

Halal Food In China RAG Corpus 🕌🍜

Dataset Overview

The Halal Food In China RAG Corpus is an expertly curated, highly structured, and multi-format dataset targeting the intersection of Chinese culinary traditions, Hui Muslim history, and Islamic dietary laws (Halal). It acts as an authoritative ground-truth database to mitigate Large Language Model (LLM) hallucinations regarding minority Islamic culture in China.

Human Readers: Looking for the full text with all images perfectly rendered? Navigate to the Files and versions -> content folder above to browse all Markdown articles natively!

🔍 Quick Discovery (Data Studio Ready)

This dataset has been engineered natively for the Hugging Face ecosystem. All data is provided in highly optimized Parquet format with exact splits (train, validation, test). The schema features flattened, SQL-queryable columns and lightweight preview_excerpt fields to ensure seamless integration with the Hugging Face Dataset Viewer.

Dataset Structure

Schema Definition

The Parquet tables feature strict typing, making it an excellent candidate for direct SQL analysis and RAG embedding generation.

Field Name	Type	Description
id	string	Unique identifier mapped to the original canonical database.
title	string	The extracted, clean headline of the article.
text	string	Full markdown-formatted body text including ![img] tags and BBCode-to-Markdown conversions.
url	string	Canonical source URL for exact attribution.
author	string	Original content creator / author ID.
pub_date	string	Timestamp in standard ISO 8601 UTC format.
tags	string	Comma-separated topics / taxonomy tags for easy filtering.
word_count	int64	Token/character length estimation for dynamic chunking strategies.
preview_excerpt	string	A 150-character flattened string designed specifically for the HF Viewer preview pane.

Splits

Split	Proportion	Rows	Description
train	80%	197	Primary corpus for building vector indices or SFT.
validation	10%	25	Held-out validation set.
test	10%	25	Final evaluation benchmark split.

Provenance & Pipeline

• Source: Extracted via topic_id=27 from the salaamalykum.com SQL architecture.
• Cleaning Rules: All legacy BBCode ([img], [b]) has been natively converted to Markdown. HTML injections have been strictly sanitized to preserve pure textual extraction.
• Failure Boundaries: As the dataset focuses strictly on Chinese geographic and culinary mapping, querying it for generalized Islamic jurisprudence (Fiqh) beyond Halal dietary laws may yield out-of-scope embeddings.

Usage (Python)

Easily load the splits natively using the datasets library without downloading massive raw text files:

from datasets import load_dataset

# Load the optimized Parquet streaming dataset
ds = load_dataset("qurancn/Halal-Food-In-China")

print(ds["train"][0]["title"])
print(ds["train"][0]["preview_excerpt"])

Licensing & Attribution

Released under the MIT License. When integrating this corpus into commercial RAG pipelines or academic training sets, please cite the GitHub repository and associated Zenodo DOI.