README.md

GERNERMED++

About

GERNERMED++ is the successor to GERNERMED, an open neural named-entity-recognition (NER) model for German texts in medical natural language processing (NLP).

Key features:

• Supported labels: Drug, Strength, Frequency, Duration, Form, Dosage
• Improved word alignment, including improved tokenization for Pharao alignment
• Introduction of transfer-learning for NER parsing
• Open, public access to models

Published paper: Our published paper is available at https://doi.org/10.1016/j.jbi.2023.104513

Online Demo: A demo page is available: Demo; and HuggingFace (see Models section)

NER demonstration:

Models

The pretrained models can be retrieved from the following URLs:

• GottBERT-based: model link
• GermanBERT-based: model link
• SpaCy (slim)-based: model link

The models are also available on the HuggingFace platform:

• GottBERT-based: HuggingFace link
• GermanBERT-based: HuggingFace link
• SpaCy (slim)-based: HuggingFace link

Scores

Note: Metric scores are evaluated by token-wise classification.

Out of Distribution Dataset (provided in OoD-dataset_GoldStandard.jsonl):

Model	Metric	Drug	Str	Freq	Dur	Form	Dos	Total
GermanBERT	Pr	0.830	0.955	0.456	1.000	0.909	0.077	0.817
	Re	1.000	0.832	0.667	0.800	0.526	0.250	0.797
	F1	0.907	0.889	0.542	0.889	0.667	0.118	0.794
GottBERT	Pr	0.872	0.868	0.933	1.000	1.000	0.125	0.882
	Re	0.932	0.980	0.718	0.400	0.684	0.250	0.868
	F1	0.901	0.921	0.812	0.571	0.813	0.167	0.865
SpaCy Slim	Pr	0.690	0.951	0.486	0.000	1.000	0.111	0.778
	Re	0.659	0.772	0.462	0.000	0.316	0.250	0.623
	F1	0.674	0.852	0.474	0.000	0.480	0.154	0.679

Test Set:

Model	Metric	Drug	Str	Freq	Dur	Form	Dos	Total
GermanBERT	Pr	0.968	0.944	0.859	0.791	0.956	0.963	0.932
	Re	0.933	0.973	0.924	0.825	0.962	0.971	0.947
	F1	0.950	0.959	0.890	0.807	0.959	0.967	0.939
GottBERT	Pr	0.966	0.969	0.879	0.813	0.949	0.966	0.941
	Re	0.926	0.965	0.951	0.825	0.972	0.972	0.952
	F1	0.946	0.967	0.914	0.819	0.961	0.969	0.946
SpaCy Slim	Pr	0.929	0.965	0.855	0.825	0.965	0.958	0.926
	Re	0.885	0.967	0.966	0.758	0.950	0.971	0.941
	F1	0.906	0.966	0.908	0.790	0.957	0.964	0.933

Setup and Usage

The models are based on SpaCy. The sample code is written in Python.

model_link="https://myweb.rz.uni-augsburg.de/~freijoha/GERNERMEDpp/GERNERMEDpp_GottBERT.zip"

# [Optional] Create env
python3 -m venv env
source ./env/bin/activate

# Install dependencies
python3 -m pip install -r requirements.txt

# Download & extract model
wget -O model.zip "$model_link"
unzip model.zip -d "model"

# Run script
python3 GERNERMEDpp.py

Citation

Cite the published paper from https://doi.org/10.1016/j.jbi.2023.104513

@article{FREI2023104513,
 title = {GERNERMED++: Semantic annotation in German medical NLP through transfer-learning, translation and word alignment},
 journal = {Journal of Biomedical Informatics},
 volume = {147},
 pages = {104513},
 year = {2023},
 issn = {1532-0464},
 doi = {https://doi.org/10.1016/j.jbi.2023.104513},
 url = {https://www.sciencedirect.com/science/article/pii/S1532046423002344},
 author = {Johann Frei and Ludwig Frei-Stuber and Frank Kramer},
 keywords = {Natural language processing, Medical NLP, Medical named entity recognition, Transfer learning, German NLP, Artificial intelligence},
}

Our ArXiv pre-print paper from https://arxiv.org/abs/2206.14504

@misc{https://doi.org/10.48550/arxiv.2206.14504,
 doi = {10.48550/ARXIV.2206.14504},  
 url = {https://arxiv.org/abs/2206.14504},  
 author = {Frei, Johann and Frei-Stuber, Ludwig and Kramer, Frank},  
 keywords = {Computation and Language (cs.CL), Artificial Intelligence (cs.AI), Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences},  
 title = {GERNERMED++: Transfer Learning in German Medical NLP},  
 publisher = {arXiv},  
 year = {2022},  
 copyright = {Creative Commons Attribution Non Commercial Share Alike 4.0 International}
}