Model Card for medical-ner-koelectra

Model Summary

This model is a fine-tuned version of the monologg/koelectra-base-v3-discriminator.

We fine-tuned the model using the KBMC and Naver X Changwon Univ NER dataset datasets.

Model Details

Model Description

• Developed by: Sungjoo Byun (Grace Byun)
• Language(s) (NLP): Korean
• License: Apache 2.0
• Finetuned from model: monologg/koelectra-base-v3-discriminator

Training Data

The model was trained using the dataset Naver X Changwon Univ NER dataset and Korean Bio-Medical Corpus (KBMC).

Model Performance

Overall Metrics

• F1 Score: 0.8886
• Loss: 0.2949
• Precision: 0.8844
• Recall: 0.8928

Class-wise Performance

Class	Precision	Recall	F1-Score	Support
AFW	0.6676	0.6326	0.6496	362
ANM	0.7476	0.7800	0.7635	600
Body	0.9731	0.9813	0.9772	1068
CVL	0.8492	0.8579	0.8536	4977
DAT	0.9078	0.9286	0.9181	2130
Disease	0.9738	0.9872	0.9805	2109
EVT	0.7332	0.7446	0.7389	1026
FLD	0.6138	0.6170	0.6154	188
LOC	0.8721	0.8691	0.8706	1734
MAT	0.5385	0.5000	0.5185	14
NUM	0.9227	0.9305	0.9266	4660
ORG	0.8917	0.8866	0.8892	3307
PER	0.8918	0.9049	0.8983	3626
PLT	0.2941	0.2174	0.2500	23
TIM	0.8644	0.9173	0.8901	278
Treatment	0.9468	0.9852	0.9656	271

Averages

Metric	Micro Avg	Macro Avg	Weighted Avg
Precision	0.8844	0.7930	0.8841
Recall	0.8928	0.7963	0.8928
F1-Score	0.8886	0.7941	0.8884

Citations

Please cite our KBMC paper:

@misc{byun2024korean,
      title={Korean Bio-Medical Corpus (KBMC) for Medical Named Entity Recognition}, 
      author={Sungjoo Byun and Jiseung Hong and Sumin Park and Dongjun Jang and Jean Seo and Minseok Kim and Chaeyoung Oh and Hyopil Shin},
      year={2024},
      eprint={2403.16158},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Model Card Contact

For any questions or issues, please contact byunsj@snu.ac.kr.