Datasets:
The dataset viewer is not available for this split.
Error code: ResponseNotFound
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
- 3 · Tổng quan · Overview
- 4 · Cấu hình · Configs
- 5 · Lược đồ · Schema
- 6 · Embeddings · Nhúng vector
- 7 · Giảm chiều — ngữ liệu chung · Dimensionality reduction — the common corpus
- 8 · Trực quan hoá · Visualizations
- 9 · Gợi ý phân tích · Analysis pointers
- 10 · Cách dùng · How to use
- 11 · Nguồn & cách thu thập · Provenance & crawl
- 12 · Giấy phép & trích dẫn · License & citation
Bộ Pháp Điển Việt Nam — phapdien.moj.gov.vn
🇻🇳 Tóm tắt. Bộ ngữ liệu cấp Điều của Bộ Pháp Điển Việt Nam — bộ pháp điển chính thức do Bộ Tư pháp công bố. Mỗi dòng
documentslà một Điều kèm toàn văn đã chuẩn hoá, chương sở thuộc, đề mục và chủ đề. Kèm theo là vector nhúng ngữ nghĩa 4096-D (embeddings), toạ độ giảm chiều trong không gian chung ViLA (reduces), và từ điển ontology song ngữ Việt–Anh (chủ đề · đề mục · thuật ngữ).🇬🇧 One-line. Article-level corpus of the Bộ Pháp Điển — the official codification of Vietnamese law published by the Ministry of Justice — with 4096-D semantic embeddings and joint-space 2-D/3-D projections shared across the six-dataset ViLA legal corpus.
Standardised to the ViLA common-corpus layout: documents (main table) · embeddings
(Nemotron-3-Embed-8B, 4096-D) · reduces (t-SNE / UMAP coordinates in one joint space fit over
all six ViLA legal datasets). Domain extras — the bilingual ontology (ontology_topics,
ontology_subjects, ontology_glossary) and the raw portal tables (subjects, tree_nodes) —
ship alongside.
3 · Tổng quan · Overview
🇻🇳 Bộ Pháp Điển được tổ chức theo ba cấp: Chủ đề (topic) → Đề mục (subject) → Điều (article). Bộ dữ liệu này phủ toàn bộ 43 chủ đề, 202 đề mục, và corpus ~66k điều, thu thập từ cổng chính thức của Bộ Tư pháp ngày 2026-07-21. · 🇬🇧 The Vietnamese legal codification has a strict three-level ontology — topic (Chủ đề) → subject (Đề mục) → article (Điều). This dataset covers the complete ontology — 43 topics, 202 subjects, and the full article corpus — crawled from the Ministry of Justice portal on 2026-07-21.
| Chỉ số · Metric | Giá trị · Value |
|---|---|
Điều luật · Articles (documents) |
65,967 rows |
Vector nhúng · Embedding rows (embeddings) |
65,997 (429 empty dividers carry embedding_dim = 0) |
Toạ độ giảm chiều · Reduction rows (reduces) |
262,272 = 65,568 content articles × {t-SNE, UMAP} × {2-D, 3-D} |
| Đề mục · Subjects | 202 |
| Chủ đề · Topics | 43 (corpus) · 42 curated in ontology_topics |
| Ngôn ngữ · Language | Vietnamese (vi); the ontology tables add English (en) labels |
| Số mã băm nội dung khác nhau · Distinct content hashes | 64,559 |
| Tổng số ký tự / từ · Total characters / words | 80,936,862 / 17,900,064 |
| Có tiêu đề chương · With chapter heading | 99.1% |
Có chú thích nguồn / liên kết vbpl.vn · With source-note / vbpl.vn link |
100.0% / 100.0% |
| Nội dung trống (mục phân cách) · Empty content (section dividers) | 429 |
| Embedding model | nvidia/Nemotron-3-Embed-8B-BF16 (4096-D) |
| Nguồn · Source | https://phapdien.moj.gov.vn/Pages/home.aspx |
| Cơ quan công bố · Publisher | Ministry of Justice of Vietnam (Bộ Tư pháp) |
| Thu thập lúc · Crawled | 2026-07-21 (202/202 đề mục fetched without error) |
Độ dài Điều luật (ký tự) · Article length (characters): median 744, mean 1,226, p90 2,554, p99 7,348, max 376,842.
Vị trí trong ngữ liệu chung ViLA · Place in the shared ViLA corpus. phapdien contributes
65,568 of the 2,326,918 points (≈ 2.33 M) in the joint embedding space that spans all six
ViLA legal datasets (cbba, tvpl-vbpl, hdpl, phapdien, tnpl, anle). Its reduces
coordinates are directly comparable, point-for-point, with every other dataset's.
4 · Cấu hình · Configs
🇻🇳 Bộ dữ liệu phơi bày 8 config. documents là bảng chính; embeddings / reduces là các
config chuẩn ViLA; phần còn lại là bảng ontology / tham chiếu. · 🇬🇧 The dataset exposes 8
configs: the three ViLA-standard ones (documents, embeddings, reduces) plus five domain
extras.
| Config | Split | Số dòng · Rows | File(s) | Vai trò · Role |
|---|---|---|---|---|
documents (default) |
train |
65,967 | documents-*.parquet (7) |
Main article table |
embeddings |
train |
65,997 | embeddings-*.parquet (33) |
Nemotron-8B 4096-D vectors |
reduces |
train |
262,272 | reduces.parquet |
Joint-space t-SNE/UMAP 2-D+3-D coordinates |
subjects |
train |
202 | subjects.parquet |
Per-đề-mục fetch metadata |
tree_nodes |
train |
245 | tree_nodes.parquet |
Raw jstree topic/subject nodes |
ontology_topics |
train |
42 | ontology_topics.parquet |
Curated bilingual topic dictionary |
ontology_subjects |
train |
202 | ontology_subjects.parquet |
Curated bilingual subject dictionary |
ontology_glossary |
train |
116 | ontology_glossary.parquet |
Cross-cutting VI↔EN legal lexicon |
⚠️ Row-count note.
documentsmaterialises 65,967 rows whileembeddingscarries 65,997 (of which 429 are empty section-dividers withembedding_dim = 0, leaving 65,568 content vectors — exactly the rows projected inreduces). Join on keys, not row order.
Khoá nối · Join story. record_id là khoá chính đảm bảo duy nhất của documents.
article_id là mã trích dẫn dễ đọc nhưng nguồn tái sử dụng ~200 mã, nên không duy nhất
toàn cục — khi nối embeddings / reduces về đúng một Điều, hãy ghép (subject_id, article_id).
Các bảng ontology (ontology_*, tree_nodes) nối qua UUID topic_id / subject_id. · 🇬🇧
record_id is the guaranteed-unique primary key of documents. article_id is the
human-facing citation code but the source reuses ~200 codes, so it is not globally unique —
join embeddings / reduces back to a single article on (subject_id, article_id). Ontology
tables join on the topic_id / subject_id UUIDs.
5 · Lược đồ · Schema
Quy ước cột song ngữ · Bilingual column convention.
documents/subjectsare English-primary (unsuffixed = EN,_vicompanion = VI);ontology_topics/ontology_subjectsuse explicit_vi/_ensuffix pairs;ontology_glossaryuses barevi/en. Vietnamese content fields (article_title,chapter_title,content_text, …) keep unsuffixed names — the source publishes them in Vietnamese only.
documents — one row per Điều (article)
| Column | Type | Notes |
|---|---|---|
record_id |
string | Guaranteed-unique row key — sha1(subject_id | article_id | content_hash)[:16]. Primary key. |
subject_id |
string | UUID of the đề-mục this article belongs to. |
topic_id |
string | UUID of the chủ-đề (topic) the đề-mục sits under. |
topic_number |
int64 | Topic display number (1–45). |
topic_title / topic_title_vi |
string | Topic name — English / Vietnamese. |
subject_number |
int64 | Đề-mục display number within its topic. |
subject_title / subject_title_vi |
string | Đề-mục name — English / Vietnamese. |
article_id |
string | Canonical citation code, e.g. Điều 39.13.TT.70.6. Not globally unique (source reuses ~200 codes) — use record_id. |
article_anchor |
string | Raw portal <a name> anchor, prefixed with #; strip client-side to recover the raw id. Not unique; may be empty. |
article_title |
string | Full title incl. the Điều N.M.X.Y prefix. |
chapter_title |
string | Chapter heading the article sits under, if any. |
source_note_text |
string | Citation back to the originating instrument (Luật/Nghị định/Thông tư). |
source_links |
list<struct<text,href>> | Hyperlinks inside source_note_text, typically to vbpl.vn. |
related_note_text |
string | Cross-references to other articles, when present. |
content_text |
string | Normalised article body (whitespace-collapsed, NBSP-stripped). |
content_char_len / content_word_count |
int64 | Character / whitespace-token counts of content_text. |
source_url |
string | Portal locator (ViewBoPD.aspx?demucid=…#<anchor>); opens the đề-mục viewer. |
scraped_at |
string | ISO-8601 UTC capture timestamp. |
embeddings — one row per article (Nemotron-8B)
Dense vectors from nvidia/Nemotron-3-Embed-8B-BF16
(4096-D, mean-pooling, passage: prefix) over content_text, plus denormalised metadata for
in-place filtering.
| Column | Type | Notes |
|---|---|---|
article_id |
string | Citation code. Join key (with subject_id) back to documents. |
embedding |
list<double> | Dense vector, 4096-D for content rows; empty dividers carry a short/zero vector with embedding_dim = 0. |
embedding_dim |
int64 | 4096 for content rows, 0 for empty dividers. |
embedding_model_id |
string | nvidia/Nemotron-3-Embed-8B-BF16. |
embedding_text_hash |
string | Hash of the embedded text (dedup / provenance). |
embedding_chunks_used |
int64 | Chunks pooled into the vector (1 when not chunked). |
embedding_chunking |
string | none (single pass) or empty (nothing to embed). |
article_title, subject_id, subject_number, subject_title, topic_id, topic_number, topic_title, chapter_title, source_url, content_char_len |
mixed | Denormalised metadata mirroring documents (*_number are strings here). |
reduces — joint-space projection coordinates (long)
One row per (article × method × dim). These are phapdien's coordinates inside one joint
t-SNE/UMAP fit over all six ViLA datasets (see §5). Long, tidy layout shared verbatim across every
ViLA card.
| Column | Type | Notes |
|---|---|---|
id |
string | article_id citation code (e.g. Điều 1.1.LQ.1). Join to documents/embeddings on article_id (+ subject_id to disambiguate). 225 codes recur (max 3×) → 65,568 rows per variant map to 65,335 distinct ids; all rows kept. |
dataset |
string | Constant phapdien — identifies the source dataset within the shared space. |
language |
string | Constant vi. |
method |
string | tsne or umap. |
dim |
int64 | 2 or 3. |
coords |
list<double> | Projected coordinates; length equals dim. |
Domain extras
subjects(202) — per-đề-mục fetch record:subject_id,topic_id,topic_number,topic_title/topic_title_vi,subject_number,subject_title/subject_title_vi,source_url,file_version,fetch_status,fetch_error,scraped_at.tree_nodes(245) — raw jstree ontology node:node_id,parent_id,kind(topic/subject),number,title,raw_text.ontology_topics(42) —topic_id,topic_number,topic_title_vi,topic_title_en,topic_note,article_count,demuc_count.ontology_subjects(202) —topic_id,topic_number,topic_title_vi,topic_title_en,subject_id,subject_title_vi,subject_title_en,article_count.ontology_glossary(116) —category(slug:instrument,structure,codification,court,agency,procedure,civil,criminal,admin,status,finance,labour,police),vi,en,note. All three ontology tables also ship as.csv;ontology.jsonbundles them as a tree.
6 · Embeddings · Nhúng vector
🇻🇳 Mỗi điều được nhúng bằng
nvidia/Nemotron-3-Embed-8B-BF16
(4096-D, mean-pooling, tiền tố passage:) từ trường content_text (tiếng Việt). · 🇬🇧 Every
article is embedded with nvidia/Nemotron-3-Embed-8B-BF16 (4096-D, mean-pooling, passage:
prefix) over the Vietnamese content_text field. Content-bearing articles get a full 4096-D
vector; the 429 empty section-dividers carry embedding_dim = 0 (filter content_char_len > 0 for
content only). A single 8-billion-parameter encoder is used across the whole ViLA corpus so vectors
are directly comparable between datasets.
from datasets import load_dataset
emb = load_dataset("tmquan/phapdien-moj-gov-vn", "embeddings", split="train")
vec = emb[0]["embedding"] # 4096-D list<double>
7 · Giảm chiều — ngữ liệu chung · Dimensionality reduction — the common corpus
🇻🇳 Config reduces là bảng dài {id, dataset, language, method, dim, coords}. Toạ độ đến từ
một lần fit chung duy nhất trên toàn bộ sáu bộ dữ liệu ViLA (2,326,918 điểm ≈ 2.33 M,
embedding nvidia/Nemotron-3-Embed-8B-BF16, KHÔNG dùng PCA). Mỗi phương pháp {tsne, umap}
được fit ở cả 2-D và 3-D; giữ tất cả các dòng. · 🇬🇧 The reduces config is a long
table {id, dataset, language, method, dim, coords}. Coordinates come from one single joint
fit over all six ViLA legal datasets (2,326,918 points ≈ 2.33 M, Nemotron-3-Embed-8B, no
PCA). Each method {tsne, umap} is fit at both 2-D and 3-D; all rows are kept.
| method | dim | rows (phapdien) |
|---|---|---|
tsne |
2 | 65,568 |
tsne |
3 | 65,568 |
umap |
2 | 65,568 |
umap |
3 | 65,568 |
| total | 262,272 |
Because the fit is joint, a phapdien article and, say, a tvpl-vbpl document that sit near each
other in coords are genuinely semantically close — cross-dataset geometry is meaningful. There are
no PCA rows and no PCA projections anywhere in this dataset.
8 · Trực quan hoá · Visualizations
Ngữ liệu chung ViLA (6 bộ) · The shared ViLA corpus (all 6 datasets)
Cùng không gian nhúng chung; phapdien là một trong sáu bộ. — The same shared embedding space;
phapdien is one of the six colours.
Cấu trúc ngữ nghĩa của phapdien trong không gian chung · This dataset's structure in the shared space
t-SNE và UMAP 2-D của phapdien, toạ độ lấy từ lần fit chung 6 bộ (không fit lại riêng), tô
màu theo chủ đề pháp luật (12 chủ đề lớn nhất + Other); kèm bản đồ cụm HDBSCAN. —
t-SNE and UMAP 2-D of phapdien, with coordinates from the shared 6-dataset joint reduce
(not a per-dataset re-fit), coloured by legal topic (top-12 + Other); plus an
HDBSCAN cluster map.
Ontology (chủ đề · đề mục · văn bản) · Ontology figures
Sơ đồ 43 chủ đề (mindmap) · 43-topic mind-map (mermaid, source in ontology_mindmap.mmd)
mindmap
root((**Bộ Pháp Điển**<br/>~66,000 Điều))
#39 Trật tự, an toàn xã hội · 5,080 Điều
#34 Thương mại, đầu tư, chứng khoán · 4,490 Điều
#14 Giao thông, vận tải · 4,267 Điều
#24 Nông nghiệp, nông thôn · 3,773 Điều
#45 Y tế, dược · 3,709 Điều
#37 Tố tụng và giải quyết tranh chấp · 3,042 Điều
#22 Ngân hàng, tiền tệ · 2,834 Điều
#33 Thuế, phí, lệ phí · 2,365 Điều
#28 Tài sản công, nợ công · 2,124 Điều
#27 Tài nguyên · 2,111 Điều
#1 An ninh quốc gia · 1,898 Điều
#25 Quốc phòng · 1,799 Điều
9 · Gợi ý phân tích · Analysis pointers
Từ toạ độ chung reduces, bạn có thể: · From the shared reduces coordinates you can:
- kNN graph / retrieval — nearest neighbours in
coords(or, for exact work, the 4096-Dembeddings) surface semantically-similar Điều withinphapdienand — thanks to the joint fit — across the other five ViLA datasets (statute text, case law, Q&A, terminology). - Clustering — run HDBSCAN / k-means on the 3-D
umapcoordinates; compare against the curatedtopic_id/subject_idlabels to measure how well the codification ontology matches embedding geometry. (A ready-made per-dataset HDBSCAN labelling is visualised above.) - Cross-dataset geometry — filter
reducesbydatasetto overlayphapdienarticles oncbbajudgments ortvpl-vbplstatutes and study where codified law sits relative to primary instruments and case law. - Topic structure — colour a 2-D projection by
topic_numberto see the 43-topic macro-structure (see §6).
10 · Cách dùng · How to use
from datasets import load_dataset
# Bảng chính · Main table (default): codified articles
docs = load_dataset("tmquan/phapdien-moj-gov-vn", "documents", split="train")
print(docs[0]["topic_title"], "/", docs[0]["topic_title_vi"])
# Vector nhúng · Embeddings (4096-D Nemotron-8B)
emb = load_dataset("tmquan/phapdien-moj-gov-vn", "embeddings", split="train")
# Toạ độ không gian chung · Shared-space coordinates (long: tsne/umap × 2D/3D)
red = load_dataset("tmquan/phapdien-moj-gov-vn", "reduces", split="train")
umap2d = red.filter(lambda r: r["method"] == "umap" and r["dim"] == 2)
# Ontology song ngữ + bảng tham chiếu · Bilingual ontology + reference tables
topics = load_dataset("tmquan/phapdien-moj-gov-vn", "ontology_topics", split="train")
subj_dic = load_dataset("tmquan/phapdien-moj-gov-vn", "ontology_subjects", split="train")
glossary = load_dataset("tmquan/phapdien-moj-gov-vn", "ontology_glossary", split="train")
subjects = load_dataset("tmquan/phapdien-moj-gov-vn", "subjects", split="train")
tree = load_dataset("tmquan/phapdien-moj-gov-vn", "tree_nodes", split="train")
documents already carries both bilingual topic/subject columns, so no join is required for
common use; join embeddings / reduces back on (subject_id, article_id).
11 · Nguồn & cách thu thập · Provenance & crawl
🇻🇳 Bộ ngữ liệu được thu thập bằng crawler tuỳ chỉnh đi qua đúng các surface mà cổng nguồn phơi ra cho trình duyệt — không có endpoint JSON / SOAP công khai. · 🇬🇧 Harvested with a custom crawler that follows the exact surfaces the source portal exposes to a browser — there is no public JSON / SOAP endpoint for the codified content:
- Fetch the topic/đề-mục tree from
TreeBoPD.aspx; the ASP.NET WebForms response embeds the entire jstree JSON (→tree_nodes,subjects). - Per đề-mục, GET
ViewBoPD.aspx?demucid=<uuid>&mapc=1to pull thefileVersiontoken. - POST to
ActionHandler.aspx(do=html) with(deMucID, fileVersion)for the full codified legal HTML. - Parse each
<p class="pDieu">into one row (attachingpChuongchapter,pGhiChusource- note + links,pChiDanrelated cross-references) →documents.
Captured at 2026-07-21T10:09:42+00:00; 202/202 đề mục fetched without error. Embeddings and
the joint reduces projection were computed downstream as part of the ViLA common-corpus build.
Ontology extras. The ontology_* tables are a hand-curated Vietnamese ↔ English dictionary
(42 topics + 202 subjects + 116 glossary terms) and the source-of-truth for the bilingual labels
denormalised into documents. ontology.json bundles all three as a tree; parallel .csv files
ship too.
Hạn chế đã biết · Known limitations
(i) The portal renders text client-side, so source_url opens the đề-mục viewer rather than
scrolling to the article — use the vbpl.vn link in source_links as the browsable source of
record. (ii) article_id is not unique (source reuses ~200 codes) — use record_id as the
primary key and join embeddings/reduces on (subject_id, article_id). (iii) Row-count skew:
documents has 65,967 rows while embeddings has 65,997 — join on keys, not row order. (iv) 429
empty divider rows carry embedding_dim = 0 and are excluded from reduces; filter
content_char_len > 0 for content-bearing articles only.
12 · Giấy phép & trích dẫn · License & citation
Văn bản pháp điển do Bộ Tư pháp công bố trên cổng công cộng, không thu phí. Bản phân phối lại này dùng giấy phép CC-BY-4.0 — vui lòng kiểm tra điều khoản của trang nguồn trước khi tái phân phối thương mại và ghi nhận tác giả gốc. · The codified text is published by the Ministry of Justice on a public portal without paywall. This redistribution is shared under CC-BY-4.0 — please check the source-website terms of use before commercial redistribution and attribute the original publisher.
- Site: https://phapdien.moj.gov.vn/Pages/home.aspx · Publisher: Ministry of Justice of Vietnam (Bộ Tư pháp) · Cross-references resolve to the National Legal Database at https://vbpl.vn/.
@misc{phapdien_2026,
title = {Vietnamese Codified Law Corpus (phapdien.moj.gov.vn)},
author = {TMQuan},
year = {2026},
howpublished = {\url{https://huggingface.co/datasets/tmquan/phapdien-moj-gov-vn}},
note = {Article-level mirror of the Vietnamese codified legal corpus, with Nemotron-8B embeddings, joint-space ViLA projections, and a curated Vietnamese--English ontology.}
}
@misc{phapdien_moj_2026,
title = {Vietnamese Codified Law Corpus},
author = {{Bộ Pháp Điển Việt Nam}},
year = {2026},
howpublished = {\url{https://phapdien.moj.gov.vn/}},
note = {Official codified body of Vietnamese law, published by the Ministry of Justice (Bộ Tư pháp).}
}
- Downloads last month
- 888










