Datasets:

YuITC
/

Vietnamese-legal-documents

license: mit
task_categories:
  - question-answering
  - sentence-similarity
  - text-generation
language:
  - vi
tags:
  - sentence-similarity
  - sentence-transformers
  - transformers
  - PyTorch
size_categories:
  - 100K<n<1M

Vietnamese Legal Documents Dataset

1. Dataset Summary

Raw data: tmnam20/BKAI-Legal-Retrieval

The Vietnamese Legal Documents Dataset is a benchmark/dataset consists of a corpus of legal texts, and train/test splits containing queries and their relevant documents.

The dataset is designed to support research and development of Information Retrieval (IR) systems, Retrieval-Augmented Generation (RAG) pipelines, and AI for LegalTech applications in Vietnamese.

2. Supported Tasks and Leaderboards

Information Retrieval (IR): Find relevant legal documents given a natural language query.
Question Answering (QA) over Documents: Used in combination with LLMs to generate answers grounded in retrieved legal documents.
Legal NLP Research: Evaluate models in the legal Vietnamese domain.

3. Dataset Structure

corpus.parquet:

column	type	description
`cid`	int	Unique document ID.
`text`	string	Full text of a legal document (Vietnamese)

train.parquet/test.parquet:

column	type	description
`question`	string	Natural-language legal query in Vietnamese.
`context_list`	list	List of text passages mapped from `cid` via the corpus (same order).
`qid`	int	Unique query ID.
`cid`	list	List of relevant document IDs (subset of `corpus.cid`).

4. Example Usuage:

from datasets import load_dataset

ds = load_dataset("YuITC/Vietnamese-Legal-Doc-Retrieval-Data")

print(ds["train"][0])
# {
#   "qid": 72600,
#   "question": "Liên đoàn Luật sư Việt Nam là tổ chức ...",
#   "context_list": ["Điều 2. Địa vị pháp lý của Liên đoàn Luật sư Việt Nam\n1. Liên đoàn Luật sư ..."]
#   "cid": [142820]
# }

5. Limitations

The dataset does not cover the entire Vietnamese legal system.
Legal texts may contain domain-specific terminology that is difficult for non-experts to interpret.