| Property | Value |
|---|---|
| Framework | Flair/PyTorch |
| Task Type | Named Entity Recognition |
| Language | German |
| Performance | 96.35% F1-Score |
What is ner-german-legal?
ner-german-legal is a specialized Named Entity Recognition model designed specifically for German legal texts. Built using the Flair framework, it can identify and classify 19 different types of legal entities, making it an invaluable tool for legal document analysis and processing.
Implementation Details
The model implements a sophisticated architecture combining Flair embeddings with an LSTM-CRF (Long Short-Term Memory - Conditional Random Field) approach. It utilizes stacked embeddings including German GloVe embeddings and bidirectional Flair embeddings (forward and backward) to achieve state-of-the-art performance.
- Trained on the LER German dataset
- Uses stacked embeddings (Word + Flair)
- Hidden size of 256 units
- Trained for 150 epochs
Core Capabilities
- Identifies 19 distinct legal entity types including courts (GRT), laws (GS), persons (PER), and organizations (ORG)
- Specialized in handling German legal terminology and structures
- Capable of processing complex legal references and citations
- Achieves high accuracy with 96.35% F1-score on benchmark datasets
Frequently Asked Questions
Q: What makes this model unique?
This model stands out due to its specialized focus on German legal texts and its comprehensive coverage of 19 different legal entity types. Its high F1-score of 96.35% makes it particularly reliable for professional legal document analysis.
Q: What are the recommended use cases?
The model is ideal for legal document processing, automated legal research, compliance checking, and legal document analysis systems. It's particularly useful for identifying references to laws, court decisions, and legal entities in German legal texts.