Upload README.md with huggingface_hub

README.md

@@ -1,73 +1,62 @@
 ---
-license: apache-2.0
+license: cc-by-4.0
+task_categories:
+- text-generation
+- question-answering
 language:
 - id
 tags:
 - legal
 - indonesia
-- regulations
-- knowledge-graph
 - rag
-task_categories:
-- text-retrieval
-- question-answering
-size_categories:
-- 100K<n<1M
+- markdown
+- regulation
+- structural-parsing
 ---
 
-# Indonesian Legal RAG Processed Database
+# 📑 Indonesian Regulation Markdown RAG Dataset (ID_REG_MD_RAG)
 
-This repository contains a fully preprocessed Indonesian legal regulation database ready for RAG (Retrieval Augmented Generation) systems.
+This repository contains a highly structured, **Markdown-optimized** collection of Indonesian Regulations (Peraturan Perundang-undangan). This dataset is specifically engineered to solve the "structure loss" problem often encountered when building Retrieval-Augmented Generation (RAG) systems for complex legal documents. 🏛️
 
-## Dataset Information
-- **Total Records**: 199,994
-- **Processing Date**: 2025-08-23
-- **Embedding Model**: Qwen/Qwen3-Embedding-0.6B
-- **Embedding Dimension**: 1024
+## 💡 The Concept: Structural Integrity for RAG
 
-## Files Description
+Legal documents in Indonesia follow a strict hierarchy: **Undang-Undang → Peraturan Pemerintah → Peraturan Menteri**, subdivided into **Bab (Chapters), Bagian (Parts), Paragraf (Paragraphs), and Pasal (Articles)**. 
 
-### Main Database
-- `processed_legal_database.parquet` - Complete preprocessed database with all features
-- `embeddings.npy` - Pre-computed embeddings array for fast loading
+Standard text-stripping methods lose this hierarchy, causing RAG systems to retrieve "Pasal 1" without knowing which Chapter or Regulation it belongs to. This dataset solves that by:
+* 🏗️ **Markdown Conversion:** All regulations are converted into clean Markdown, using headers (`#`, `##`, `###`) to preserve the legal hierarchy.
+* 🔗 **Contextual Anchoring:** Each chunk maintains its structural relationship to the parent heading, allowing LLMs to "see" the full path of a regulation.
+* ✂️ **RAG-Ready Chunking:** Optimized for semantic splitters that respect Markdown boundaries, ensuring that an Article is never cut in half during retrieval.
 
-### Supporting Components  
-- `tfidf_components.pkl` - TF-IDF vectorizer and matrix for keyword matching
-- `knowledge_graph.json` - Legal concept relationships and hierarchy
-- `config.json` - Processing configuration and parameters
-- `processing_summary.json` - Processing statistics and metadata
+## 🛠️ Data Acquisition & Pipeline
 
-### Utilities
-- `database_loader.py` - Helper class for easy database loading
+This dataset was compiled using a systematic developer workflow:
+* **Source:** Official Indonesian regulatory repositories and government gazettes. 🌐
+* **Transformation:** Raw documents (PDF) were processed through a custom pipeline to extract structural elements and reformat them into consistent Markdown syntax. 🔄
+* **Methodology:** Built using legitimate developer-sanctioned methods, ensuring clean data lineage and high-fidelity text extraction without the noise of typical web scraping. ✅
 
-## Features Included
+## 🧪 Use Cases
 
-Each record contains:
-- **Basic Information**: Regulation type, number, year, enacting body, content
-- **Authority Score**: Hierarchical legal authority level (0-1)  
-- **Temporal Score**: Regulation currency/relevance (0-1)
-- **Legal Richness**: Legal term density and complexity (0-1)
-- **Cross-Reference Strength**: Inter-regulation connectivity (0-1)
-- **Completeness Score**: Information completeness (0-1)
-- **Knowledge Graph Features**: Entity extraction and concept relationships
-- **Pre-computed Embeddings**: Ready-to-use vector representations
+* **Advanced RAG Pipelines:** Powering legal assistants that can accurately identify the specific Bab or Bagian a statute belongs to. 🔍
+* **Semantic Search:** Building search engines that understand the hierarchical importance of legal headers. 🚀
+* **Legal Knowledge Graphs:** Using the Markdown structure to map relationships between different regulatory tiers. 🧠
+* **LLM Fine-Tuning:** Training models to generate legal documents that follow the official Indonesian legislative format. 🎓
 
-## Usage
+## 📂 Dataset Structure
 
-```python
-from database_loader import ProcessedLegalDatabaseLoader
+Each entry is optimized for ingestion:
+* **filename:** The official name of the regulation (e.g., UU Nomor 1 Tahun 2024).
+* **content_md:** The full text of the regulation in structured Markdown format. 📑
+* **metadata:** Includes the type of regulation, year of issuance, and original source references.
 
-# Load complete database
-loader = ProcessedLegalDatabaseLoader(".", local_mode=True)
-df, embeddings, tfidf, kg, config = loader.load_all()
+## 📜 License & Attribution
 
-# Ready for RAG system - no preprocessing needed!
-```
+This dataset is licensed under the **Creative Commons Attribution 4.0 International (CC BY 4.0)**. 📝
 
-## Memory Requirements
-- **Database**: ~1260MB
-- **Embeddings**: ~781MB  
-- **Total**: ~2747MB
+### Attribution
+1. **Dataset Curator:** Azzindani (via Hugging Face Datasets).
+2. **Original Source:** Government of the Republic of Indonesia (Public Domain).
 
-## License
-Please ensure compliance with Indonesian legal data usage regulations.
+---
+**Disclaimer:** *This dataset is provided for research and development purposes. While formatted for high accuracy in AI applications, users should always verify the text against official government gazettes (Lembaran Negara) for formal legal activities.* ⚠️
+
+---