---
language:
- vi
- en
license: cc-by-nc-4.0
size_categories:
- 1K
— the official Vietnam Cartographic Publishing House atlas of the
**post-merger administrative units** introduced by **National Assembly
Resolution 202/2025/QH15 of 12 June 2025** and the **34 follow-up
Standing Committee resolutions of 16 June 2025**.
The merger collapsed Vietnam from **63 first-level units to 34** (28
provinces + 6 centrally-administered cities) and re-drew the second
tier from **705 districts ÷ 10,599 communes** down to **3,321 communes /
wards / special administrative units**. This dataset captures every
surviving entity together with its merger lineage, area, population,
administrative centre, decree of authority, and a polygon (or point)
geometry.
## At a glance
| Stat | Value |
| ------------------------------------- | ----------------- |
| First-level admin units (post-merger) | **34** |
| Second-level admin units (post-merger)| **3,321** |
| People's-committee headquarters | **3,357** |
| Total Vietnam population (2024) | **113,571,926** |
| Total Vietnam land area (km²) | **331,325.62** |
| Province polygons (GeoJSON) | **34** |
| Commune polygons (GeoJSON) | **3,321** |
| Max merger fanout (predecessor units) | **16** |
## What's on the Hub
```
.
├── README.md (this file)
├── _stats.json numbers / tables this card quotes
├── data/
│ ├── all.parquet 6,712 rows = provinces + communes + committees
│ ├── provinces.parquet 34 rows
│ ├── communes.parquet 3,321 rows
│ └── committees.parquet 3,357 rows
├── geo/
│ ├── provinces.geojson 34 polygons (FeatureCollection)
│ └── communes.geojson 3,321 polygons (FeatureCollection)
├── reduced/
│ └── reduced.parquet UMAP 2-D coords + HDBSCAN cluster id
├── figures/analysis/ 14 PNG + 14 HTML interactive Plotly figures
├── notebooks/DATAANALYSIS.ipynb
├── docs/{DATAPROCESSING,DATAANALYSIS}.md
└── raw/{admin_units,committees}.json (source listings)
```
Every figure under ``figures/`` ships in **both formats** — the static
PNG for inline rendering in this dataset card, plus a self-contained
**interactive HTML** with the full Plotly toolkit (pan / zoom / hover
tooltips). Click any HTML link or download the ``.html`` to open it
locally in a browser.
## Per-macro-region inventory
| Macro-region (EN) | Provinces | Communes | Committees |
| ------------------------------------------ | --------: | -------: | ---------: |
| Northern Midlands and Mountain Areas | 10 | 841 | 842 |
| Red River Delta | 5 | 527 | 502 |
| North Central and Central Coastal Areas | 8 | 738 | 737 |
| Central Highlands | 3 | 361 | 380 |
| Southeast | 3 | 359 | 358 |
| Mekong River Delta | 5 | 495 | 537 |## Curator pipeline
The on-disk shape under this repo is produced by the same five-stage
NeMo Curator pipeline as `personas-vn`:
```
download → parse → extract → embed → reduce
```
* **download** — POST to four endpoints exposed by ``sapnhap.bando.com.vn``
(``p.co_dvhc``, ``p.co_uyban``, ``p.co_dvhc_id``, ``pread_json``);
~6,700 calls; cached to disk so re-runs are free.
* **parse** — normalise Vietnamese-formatted numbers (``"4.199.824"`` →
4 199 824), summarise GeoJSON to centroid + bbox + WKT, attach
parent-province for every commune & committee.
* **extract** — TF-IDF keyword extraction over the merger-lineage
prose; macro-region attachment from the curated 34 → 6 GSO mapping.
* **embed** — ``sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2``
on CPU (384-d) over the canonical Vietnamese descriptors.
* **reduce** — UMAP → 2-D coords + density-based HDBSCAN clusters.
See [`docs/DATAPROCESSING.md`](docs/DATAPROCESSING.md) for the full
crawl protocol, [`docs/DATAANALYSIS.md`](docs/DATAANALYSIS.md) for the
analytical walkthrough, and the [`notebooks/DATAANALYSIS.ipynb`](notebooks/DATAANALYSIS.ipynb)
notebook for the executed Plotly outputs.
## Usage
```python
from datasets import load_dataset
# Default config (all 6,712 rows)
ds = load_dataset("REPO_ID")["train"]
print(ds.column_names)
# -> ['id', 'kind', 'ma', 'ten', 'type', 'ten_short', 'area_km2',
# 'population', 'density', 'capital', 'address', 'phone', 'decree',
# 'decree_url', 'predecessors', 'parent_ma', 'parent_ten',
# 'centroid_lon', 'centroid_lat', 'bbox', 'geom_type', 'wkt',
# 'predecessors_list', 'n_predecessors', 'macro_region',
# 'embed_text', 'keywords']
# Just the 34 first-level units
provinces = load_dataset("REPO_ID", "provinces")["train"]
# GeoJSON (download separately — datasets doesn't load .geojson)
import huggingface_hub as hf, json
path = hf.hf_hub_download("REPO_ID", "geo/provinces.geojson",
repo_type="dataset")
fc = json.loads(open(path, "r", encoding="utf-8").read())
print(len(fc["features"]))
```
## Citation
The underlying data belongs to the **Ministry of Agriculture and
Environment** and the **Vietnam Cartographic Publishing House**
(Nhà Xuất Bản Tài Nguyên - Môi Trường và Bản Đồ Việt Nam):
* ISBN: 978-632-622-303-3
* Publication ID: 1027-2026/CXBIPH/03-129/BĐ
* Published: 2026 (Quyết định số 30/QĐ-NXBTNMT, 16 April 2026)
* Source: ·
Authoritative legal sources for the merger:
* National Assembly Resolution 202/2025/QH15 (12 June 2025)
* Standing Committee resolutions of 16 June 2025 (×34)
* Government decrees published at
If this mirror or the analytical figures are useful in academic work,
please credit the publishers above.