dienmoc/vietnamese-legal-documents

Vietnamese Legal Documents

A comprehensive dataset of 518,255 Vietnamese legal documents sourced from thuvienphapluat.vn — the largest Vietnamese legal document repository. The dataset covers laws, decrees, circulars, decisions, and other official documents issued by Vietnamese government bodies, spanning from 1924 to 2026.

---

At a Glance

🗂️ Total documents	518,255
📅 Date range	1924 – 2026
🏛️ Issuing authorities	1,335 unique bodies
📋 Document types	36 unique types
🌐 Language	Vietnamese
💾 Content size	~3.6 GB (parquet)

---

Dataset Structure

This dataset is split into two configs to allow fast metadata access without loading the full text.

Config	Split	Rows	Size	Description
metadata	data	518,255	~82 MB	9 metadata columns, no text content
content	data	518,255	~3.6 GB	id + full markdown document text

Join on the id column to get both metadata and content.

---

Load the Dataset

from datasets import load_dataset

# Load metadata only (fast, ~82 MB)
ds = load_dataset("th1nhng0/vietnamese-legal-documents", "metadata")
df = ds["data"].to_pandas()
print(df.head())

# Load full text content (~3.6 GB)
ds_content = load_dataset("th1nhng0/vietnamese-legal-documents", "content")

# Join metadata + content
import pandas as pd
meta = load_dataset("th1nhng0/vietnamese-legal-documents", "metadata")["data"].to_pandas()
text = load_dataset("th1nhng0/vietnamese-legal-documents", "content")["data"].to_pandas()
df = meta.merge(text, on="id")
print(df.columns.tolist())

---

Schema

metadata config

Column	Type	Description
id	int64	Unique numeric document ID
document_number	string	Official document number (e.g. 115/NQ-HĐBCQG)
title	string	Full Vietnamese title
url	string	Source URL on thuvienphapluat.vn
legal_type	string	Document type (Quyết định, Công văn, Nghị quyết, …)
legal_sectors	string	Pipe-separated sector/topic tags
issuing_authority	string	Name of the issuing government body
issuance_date	string	Issue date in DD/MM/YYYY format
signers	string	Pipe-separated name:id pairs of signatories

content config

Column	Type	Description
id	int64	Document ID — join key with the metadata config
content	string	Full document text converted to Markdown

---

Statistics

Documents by Year

Top 15 Document Types

Top 15 Legal Sectors

---

Use Cases

• 🔍 Legal information retrieval — build search engines over Vietnamese law
• 🤖 LLM fine-tuning — train or fine-tune language models on legal Vietnamese
• 📊 Legal NLP research — NER, classification, summarization, QA
• 📈 Policy analysis — track legislative trends over time
• 🌏 Low-resource NLP — Vietnamese legal text is underrepresented in existing datasets

---

Data Collection

This is an independent personal research project. Documents were collected from thuvienphapluat.vn — a public legal document portal — via their sitemap. This project has no affiliation with thuvienphapluat.vn.

HTML content was converted to Markdown using BeautifulSoup. Only Vietnamese-language documents were retained; English versions and technical standards (Tiêu chuẩn) were excluded.

---

License & Legal Basis

Vietnamese legal documents (laws, decrees, circulars, decisions, and other normative acts) are public domain by Vietnamese law. Under the Law on Access to Information (Luật Tiếp cận thông tin, No. 104/2016/QH13) and the Law on Promulgation of Legal Documents (No. 64/2025/QH15), official legal normative documents issued by state agencies must be made publicly accessible free of charge.

The compiled dataset (collection, processing, metadata schema, and Markdown conversion) is released under CC BY 4.0.

Intended for research purposes only.

---

Citation

@dataset{ngo_thinh_2026_vietnamese_legal,
  title        = {Vietnamese Legal Documents},
  author       = {Ngô, Thịnh},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/th1nhng0/vietnamese-legal-documents},
  note         = {518,255 Vietnamese legal documents compiled for research purposes}
}