|
--- |
|
license: cc-by-2.0 |
|
language: |
|
- en |
|
tags: |
|
- pico |
|
- pubmed |
|
- ner |
|
- pubmedbert |
|
--- |
|
|
|
# PubMed PICO extraction (NER model) |
|
|
|
This is a fine-tuned multi-class NER model for PICO labels within PubMed biomedical database abstracts, trained with [NERDA](https://ebanalyse.github.io/NERDA/). |
|
|
|
The mode predicts: |
|
|
|
| Class | **F1-Score** | Precision | Recall | |
|
|
|
|B-Patient | **0.72** | 0.69 | 0.76 |
|
|
|
|I-Patient | **0.77** | 0.74 | 0.81 |
|
|
|
|B-Intervention | **0.62** | 0.61 | 0.63 | |
|
|
|
|I-Intervention | **0.70** | 0.70 | 0.71 | |
|
|
|
|B-Control | **0.59** | 0.56 | 0.63 | |
|
|
|
|I-Control | **0.62** | 0.64 | 0.61 | |
|
|
|
|B-Outcome | **0.49** | 0.50 | 0.49 | |
|
|
|
|I-Outcome | **0.56** | 0.63 | 0.51 | |
|
|
