Azzindani/ID_REG_MD_Embed

📧 Github 🔗 LinkedIn

🏛️ Indonesian Legal RAG Dataset with Embeddings & TF-IDF Vectors

📖 What is this dataset?

This dataset contains Indonesian legal documents with pre-computed embeddings and TF-IDF vectors for building RAG (Retrieval-Augmented Generation) systems. It includes regulations, laws, and legal documents from Indonesia with ready-to-use vector representations.

🎯 Perfect for: Legal AI, Indonesian NLP, RAG systems, semantic search, and legal chatbots

✨ Key Features

• 📚 ~10,000 Indonesian regulations
• 🎯 Dense embeddings (1024-dimensional vectors)
• 📊 TF-IDF vectors (5000-dimensional sparse vectors)
• ⚖️ Complete legal metadata (regulation type, year, authority, etc.)
• 🔍 Ready for semantic search - no preprocessing needed!
• 🚀 Optimized for RAG retrieval systems
• 📜 Full document text included
• 🏛️ Indonesian legal hierarchy preserved

📊 What's Inside?

📋 Document Fields

• Document content: Full text of legal documents
• Metadata: Regulation type, number, year, issuing body
• Structure: Chapters, articles, and sections
• Embeddings: 1024-dim dense vectors for semantic similarity
• TF-IDF: 5000-dim sparse vectors for keyword matching

🏛️ Legal Document Types

• 📜 UUD - Constitution (Undang-Undang Dasar)
• 📋 UU - Laws (Undang-Undang)
• 🏛️ PP - Government Regulations (Peraturan Pemerintah)
• 🎯 PERPRES - Presidential Regulations (Peraturan Presiden)
• 📑 PERMEN - Ministerial Regulations (Peraturan Menteri)
• 🏙️ PERDA - Regional Regulations (Peraturan Daerah)

📋 Complete Data Schema

Field	Type	Dimension	Description	Example
🆔 global_id	string	-	Unique document identifier	"legal_doc_001"
🏷️ local_id	string	-	Source-specific ID	"local_001"
📜 regulation_type	string	-	Type of regulation	"UU", "PP", "PERPRES"
🏛️ enacting_body	string	-	Government entity	"Kementerian Hukum"
🔢 regulation_number	string	-	Official number	"No. 8 Tahun 1999"
📅 year	string	-	Year enacted	"1999"
📝 about	string	-	Document topic	"Perlindungan Konsumen"
📅 effective_date	string	-	When it takes effect	"1999-04-20"
📖 chapter	string	-	Document chapter	"BAB I"
📄 article	string	-	Specific article	"Pasal 1"
📄 content	string	-	Full document text	"Pasal 1. Dalam Undang-undang..."
🎯 embedding	list[float]	1024	Dense semantic vectors	[0.1, 0.2, -0.3, ...]
📊 tfidf_vector	list[float]	5000	Sparse keyword vectors	[0.0, 0.15, 0.0, 0.8, ...]
🔢 chunk_id	int	-	Document chunk number	1

⚠️ Important Notes

✅ What This Dataset Is Good For:

• 🤖 Building Indonesian legal RAG systems
• 🔍 Semantic search in legal documents
• 📊 Legal document clustering and analysis
• 🎯 Training legal AI models
• 📚 Legal research and information retrieval

❌ What This Dataset Is NOT:

• ⚖️ Not legal advice - for research/education only!
• 🎯 Not 100% accurate - always verify with official sources
• 🌍 Not for other legal systems - specific to Indonesia
• ⏰ Not always current - laws change over time
• 🔍 Not query-ready - you need to encode your queries

🚨 Limitations:

• 📏 Fixed dimensions - embeddings are 1024-dim, TF-IDF is 5000-dim
• 🎯 Model dependency - embeddings quality depends on the model used
• 📊 TF-IDF vocabulary - limited to 5000 most common terms
• ⏰ Point-in-time - vectors represent documents at creation time
• 🔍 No query encoder included - you need your own query encoding

🔄 Preprocessing Pipeline

The dataset was created using this pipeline:

1. 📥 Data Collection - Indonesian legal documents gathered
2. 🧹 Text Cleaning - Remove noise, normalize formatting
3. 📊 TF-IDF Vectorization - Create 5000-dim sparse vectors
4. 🎯 Embedding Generation - Create 1024-dim dense vectors
5. ✅ Quality Control - Validate dimensions and formats
6. 💾 Dataset Creation - Package into HuggingFace format

---

🚀 Happy Legal AI Building! 🏛️✨

Ready-to-use vectors for Indonesian legal RAG systems 🎯📊