AsmaaBoudjenane/arabic-legal-qa-dataset
Domain:
natural language processingRecord type:
dataset
Creator:
Asm
Host:
Arabic Legal QA Dataset: Algerian Supreme Court Rulings and QA Pairs
# Arabic Legal QA Dataset
A dataset for Arabic legal question answering comprising Algerian Supreme Court rulings and legal QA pairs.
## Overview
This dataset was created to support research in Arabic legal NLP and retrieval-augmented generation (RAG) systems. It contains authentic court rulings from the Algerian Supreme Court and synthetic question-answer pairs for training and evaluation.
## Dataset Files
The repository contains two Excel files:
- **legal_corpus.xlsx** — Algerian Supreme Court rulings
- **qa_dataset.xlsx** — Question-answer pairs
## Dataset Statistics
| Metric | Value |
|--------|-------|
| Total court rulings | 1,248 |
| Total QA pairs | 3,742 |
| Total tokens | 463,631 |
| Vocabulary size | 36,080 |
| Average document length | 371.5 tokens |
| Median document length | 336 tokens |
| Maximum document length | 1,781 tokens |
| Average question length | 9.99 tokens |
| Average answer length | 84.82 tokens |
| Average retrieved context length | 355.91 tokens |
| Duplicate documents | 0 |
| Duplicate QA pairs | 0 |
| Missing values | 0 |
## Data Format
### Legal Corpus (`legal_corpus.xlsx`)
| Column | Description |
|--------|-------------|
| case_id | Unique identifier for the ruling |
| title | Case title |
| keywords | Legal keywords associated with the case |
| full_text | Complete ruling text in Arabic |
### QA Dataset (`qa_dataset.xlsx`)
| Column | Description |
|--------|-------------|
| question_id | Unique identifier for the QA pair |
| question | Legal question in Arabic |
| answer | Reference answer in Arabic |
| question_type | Case-specific, verdict, legal reasoning, or statutory interpretation |
| relevant_case_ids | Related court ruling IDs |
## Dataset Construction
**Court Rulings:** 1,248 rulings were collected from publicly accessible Algerian Supreme Court archives via automated web scraping, followed by HTML parsing, metadata normalization, Arabic text normalization, diacritics removal, and manual quality verificatio …