|
|
--- |
|
|
license: apache-2.0 |
|
|
--- |
|
|
|
|
|
|
|
|
# 🧾 Indonesian Legal QA Dataset |
|
|
|
|
|
This repository contains a **question-answer (QA) dataset** generated from parsed Indonesian regulations, focusing on **legal quoting and comprehension**. Designed to facilitate legal-aware LLMs, the dataset provides direct QA mappings to individual articles for contextual understanding and reference. |
|
|
|
|
|
--- |
|
|
|
|
|
## 📌 Dataset Highlights |
|
|
|
|
|
* **Source**: Generated from the [ID\_REG\_Parsed](https://huggingface.co/datasets/Azzindani/ID_REG_Parsed) repository |
|
|
* **Format**: QA pairs based on individual articles (no chunking) |
|
|
* **Scale**: Augmented by applying 10 QA templates across suitable regulation entries |
|
|
* **Filtering**: Programmatic filtering removes redundant or overly broad article explanations |
|
|
* **Target Use**: Train/test LLMs for **regulation comprehension**, **legal quoting**, and **document-level QA** |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚙️ Pipeline Overview |
|
|
|
|
|
* **Environment**: Executed in a single Jupyter Notebook on **Kaggle Cloud** |
|
|
|
|
|
* **Data Flow**: |
|
|
|
|
|
1. **Pull** parsed articles from `ID_REG_Parsed` |
|
|
2. Filter and refine results for clarity and legal context |
|
|
3. Apply **template-driven QA generation** (10 variations) |
|
|
4. **Push** QA dataset directly to this repository |
|
|
|
|
|
* **Performance**: |
|
|
|
|
|
* Completed in \~20 minutes using Kaggle GPU resources |
|
|
* Cloud-to-cloud transfer without local storage dependency |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧠 Use Cases |
|
|
|
|
|
* Fine-tuning LLMs for **legal question answering** |
|
|
* Benchmarks for **article referencing and quoting** |
|
|
* Few-shot prompting for legal search assistants |
|
|
* Legal text evaluation with grounded answers |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚠️ Disclaimer |
|
|
|
|
|
This dataset is intended for **research and development** only. QA pairs are generated synthetically from publicly available legal text and may not reflect official interpretations. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🙏 Acknowledgments |
|
|
|
|
|
* **[Hugging Face](https://huggingface.co/)** for hosting open datasets |
|
|
* **[Kaggle](https://www.kaggle.com/)** for compute and cloud-to-cloud capabilities |
|
|
|
|
|
--- |
|
|
|
|
|
|
