Deep learning with language models improves named entity recognition for PharmaCoNER

Codes and models.

The repository is organized as follows.

PharmaCoNER
  -data
    -train.txt
    -dev.txt
    -test.txt
  -src
    -vocab.txt
    -config.json
    -pytorch_model.bin
  -outputs
  -labels.txt
  -metric.py
  -utils_ner.py
  -PharmaCoNER_run.py

• src: used to provide the BERT models. The models can be downloaded from here.
• data: used to provide the processed PharmaCoNER dataset.
• outputs: used to store the output files.
• Furthermore, because the original link cannot be opened, we uploaded the original PharmaCoNER corpus for readers to study.

Tested environments

• Ubuntu 18.04
• python 3.6.10
• transformers 2.8.0
• torch 1.5.0
• numpy 1.18.3
• tqdm 4.45.0
• scipy 1.4.1
• tensorboardX 2.0

Run the model

CUDA_VISIBLE_DEVICES=0 python PharmaCoNER_run.py\
--do_train   \
--do_eval    \
--do_predict   \
--evaluate_during_training   \
--model_type bert   \
--data_dir /$YourPath/PharmaCoNER/data   \
--model_name_or_path /$YourPath/src/$BERT_model_file   \
--max_seq_length 300   \
--per_gpu_train_batch_size 16   \
--per_gpu_eval_batch_size 16   \
--save_steps 100000    \
--labels /$YourPath/PharmaCoNER/labels.txt    \
--learning_rate 2e-5   \
--num_train_epochs 20   \
--seed 9  \
--output_dir /$YourPath/PharmaCoNER/outputs/$Output_file