BioBERT-CFA-MTM

This is the implementation of Biomedical Named Entity Recognition with the Combined Feature Attention and Fully-Shared Multi-Task Learning

Requirements

• python3
• pip3 install -r requirements.txt
• Downloading BioBERT and using our vocat.txt

Run

python run_ner_multi.py --data_dir=data/all/ --bert_model=./biobert-base-cased-v1.1 --task_n ame=ner --output_dir=./output --max_seq_length=128 --do_train --num_train_epochs=10 --do_eval --warmup_proportion=0.1 -- learning_rate=5e-5 --train_batch_size=64

Example

BioBERT-MTM

NCBI-disease

             precision    recall  f1-score   support

          I     0.8789    0.8920    0.8854       537
          B     0.8987    0.9146    0.9066       960

avg / total     0.8916    0.9065    0.8990      1497