KoELECTRA-small-v3-modu-ner
| Property | Value |
|---|---|
| Parameter Count | 14.1M |
| Model Type | Token Classification |
| Architecture | ELECTRA-small |
| Performance | F1: 0.8339, Accuracy: 0.9628 |
What is KoELECTRA-small-v3-modu-ner?
KoELECTRA-small-v3-modu-ner is a specialized Korean Named Entity Recognition (NER) model fine-tuned on the monologg/koelectra-small-v3-discriminator architecture. It's designed to identify and classify 15 different types of entities in Korean text, following the Korean Information and Communications Technology Association (TTA) standards.
Implementation Details
The model employs the BIO (Begin, Inside, Outside) tagging system for entity recognition and was trained using the Korean National Institute of Korean Language's corpus. Training involved 20 epochs with a learning rate of 5e-05 and achieved impressive metrics including 82.32% precision and 84.49% recall.
- Uses Native AMP mixed precision training
- Trained with batch size of 64
- Implements linear learning rate scheduling
- Optimized using Adam optimizer
Core Capabilities
- Recognizes 15 distinct entity types including Person, Location, Organization
- Handles complex Korean text processing
- Achieves 96.28% accuracy on validation set
- Supports real-time inference through pipeline implementation
Frequently Asked Questions
Q: What makes this model unique?
The model's strength lies in its comprehensive coverage of Korean entity types and high accuracy metrics, making it particularly useful for Korean language processing tasks. Its small parameter count (14.1M) makes it efficient while maintaining strong performance.
Q: What are the recommended use cases?
The model is ideal for Korean text analysis tasks including: information extraction, content categorization, and automated text processing systems. It's particularly well-suited for applications requiring identification of named entities in Korean text.
