ClinicalNER
Model Description
This is a multilingual clinical NER model extracting DRUG, STRENGTH, FREQUENCY, DURATION, DOSAGE and FORM entities from a medical text.
It consist of XLM-R Base fine-tuned on n2c2 (English). It is the model that obtains the best results on our French evaluation test set MedNERF in a zero-shot cross-lingual transfer setting.
Evaluation Metrics on MedNERF dataset
- Loss: 0.692
- Accuracy: 0.859
- Precision: 0.817
- Recall: 0.791
- micro-F1: 0.804
- macro-F1: 0.819
Usage
from transformers import AutoModelForTokenClassification, AutoTokenizer
model = AutoModelForTokenClassification.from_pretrained("Posos/ClinicalNER")
tokenizer = AutoTokenizer.from_pretrained("Posos/ClinicalNER")
inputs = tokenizer("Take 2 pills every morning", return_tensors="pt")
outputs = model(**inputs)
Citation information
@inproceedings{mednerf,
title = "Multilingual Clinical NER: Translation or Cross-lingual Transfer?",
author = "Gaschi, Félix and Fontaine, Xavier and Rastin, Parisa and Toussaint, Yannick",
booktitle = "Proceedings of the 5th Clinical Natural Language Processing Workshop",
publisher = "Association for Computational Linguistics",
year = "2023"
}
- Downloads last month
- 209
Inference Providers
NEW
This model is not currently available via any of the supported third-party Inference Providers, and
the model is not deployed on the HF Inference API.
Dataset used to train Posos/ClinicalNER
Evaluation results
- micro-F1 score on MedNERFtest set self-reported0.804
- precision on MedNERFtest set self-reported0.817
- recall on MedNERFtest set self-reported0.791
- accuracy on MedNERFtest set self-reported0.859