| Property | Value |
|---|---|
| License | MIT |
| Paper | arXiv:2402.15343 |
| Supported Languages | English, French, German, Italian, Spanish, Portuguese, Polish, Dutch, Russian |
| Best F1 Score | 0.6231 (with two embeddings) |
What is NuNER-multilingual-v0.1?
NuNER-multilingual-v0.1 is a state-of-the-art entity recognition model built on Multilingual BERT architecture. It's specifically fine-tuned on an artificially annotated multilingual subset of the OSCAR dataset, providing domain and language-independent embeddings for entity recognition tasks. The model demonstrates impressive performance improvements over the base mBERT model, achieving a 0.6231 F1 macro score with its innovative two-embedding approach.
Implementation Details
The model is implemented using the Transformers library and PyTorch, offering flexible deployment options for both inference and fine-tuning. It uniquely combines outputs from different layers of the transformer architecture to achieve superior performance.
- Built on Multilingual BERT base architecture
- Supports both single and two-embedding approaches
- Implements token classification pipeline
- Provides pre-trained weights optimized for entity recognition
Core Capabilities
- Multilingual entity recognition across 9+ languages
- Domain-independent embedding generation
- Flexible integration with existing NLP pipelines
- Support for both inference and fine-tuning workflows
- Superior performance compared to base mBERT (0.5206 vs 0.6231 F1 score)
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its ability to generate high-quality embeddings for entity recognition across multiple languages while maintaining domain independence. The innovative two-embedding approach significantly improves performance over traditional methods.
Q: What are the recommended use cases?
The model is ideal for multilingual entity recognition tasks, cross-lingual information extraction, and as a foundation for fine-tuning on specific domain entity recognition tasks. It's particularly valuable for applications requiring robust entity recognition across multiple languages.