BC5CDR_bioBERT_NER
| Property | Value |
|---|---|
| Base Model | dmis-lab/biobert-v1.1 |
| Task | Named Entity Recognition (NER) |
| Author | judithrosell |
| Model Hub | Hugging Face |
What is BC5CDR_bioBERT_NER?
BC5CDR_bioBERT_NER is a specialized Named Entity Recognition model fine-tuned on the BioBERT architecture for identifying chemical and disease entities in biomedical text. The model demonstrates exceptional performance, particularly in chemical entity recognition with a 98% F1-score, while maintaining strong disease entity detection capabilities at 85% F1-score.
Implementation Details
The model was trained using a carefully optimized process with the following specifications: Adam optimizer with betas=(0.9,0.999), linear learning rate scheduler, and a learning rate of 2e-05. Training was conducted over 3 epochs with a batch size of 32 (16 base with 2 gradient accumulation steps).
- Architecture: Fine-tuned BioBERT v1.1 base model
- Training Duration: 3 epochs
- Validation Loss: 0.0808 (final)
- Framework Compatibility: Transformers 4.35.2, PyTorch 2.1.0+cu121
Core Capabilities
- Chemical Entity Recognition: 99% precision, 98% recall
- Disease Entity Recognition: 83% precision, 88% recall
- Overall Micro-average: 97% F1-score
- Support Size: 110,280 entities (103,336 chemical, 6,944 disease)
Frequently Asked Questions
Q: What makes this model unique?
The model's exceptional performance in chemical entity recognition (98% F1-score) combined with strong disease entity detection makes it particularly valuable for biomedical text analysis. The balanced precision-recall trade-off demonstrates robust real-world applicability.
Q: What are the recommended use cases?
This model is ideal for biomedical text mining applications, particularly those focusing on chemical compound identification and disease mention detection. It's especially suitable for large-scale document processing where high accuracy in chemical entity recognition is crucial.
