roberta-es-clinical-trials-ner
| Property | Value |
|---|---|
| Author | lcampillos |
| License | CC BY 4.0 |
| Performance | F1: 0.8647, Accuracy: 0.9583 |
| Base Model | bsc-bio-ehr-es |
What is roberta-es-clinical-trials-ner?
This is a specialized Named Entity Recognition (NER) model designed to identify medical entities in Spanish clinical trial texts. Built upon the bsc-bio-ehr-es architecture, it recognizes four crucial UMLS semantic groups: anatomical structures (ANAT), chemical/pharmacological substances (CHEM), disorders (DISO), and medical procedures (PROC). The model achieves impressive performance metrics with an F1-score of 0.8647 and accuracy of 0.9583.
Implementation Details
The model was fine-tuned on the CT-EBM-SP corpus, comprising 1,200 clinical trial texts, including 500 journal abstracts and 700 clinical trial announcements. Training utilized Adam optimizer with a learning rate of 2e-05 over 4 epochs, achieving optimal performance with minimal environmental impact (0.01 kg CO2 emissions).
- Training batch size: 16 with linear learning rate scheduling
- Built using Transformers 4.17.0 and PyTorch 1.10.2
- Validated on diverse medical entity types with strong per-class performance
Core Capabilities
- Accurate detection of anatomical terms (67.83% F1)
- Excellent recognition of chemical substances (91.95% F1)
- Strong performance on disorder identification (88.61% F1)
- Reliable procedure detection (84.50% F1)
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in Spanish medical text analysis, particularly for clinical trials, with comprehensive coverage of four critical UMLS semantic groups. Its high accuracy and specialized training make it valuable for medical text processing tasks.
Q: What are the recommended use cases?
The model is designed for analyzing Spanish clinical trial documents, research abstracts, and medical announcements. However, it should not be used for medical decision-making without human supervision, as stated in its usage guidelines.
