AICUP 2023 - Electronic Medical Record PHI Extraction

Environment

• OS: Ubuntu 20.04
• Python 3.8.5
• GPU: RTX 3060

Install

We use Poetry to manage dependencies.

poetry init
poetry install

We also provide a requirements.txt file for those who prefer to use pip.

pip install -r requirements.txt

Data Preparation

Put the data in the data folder.

data
├── First_Phase_Release
│   ├── First_Phase_Text_Dataset
│   ├── Validation_Release
│   └── answer.txt
├── Second_Phase_Dataset
│   ├── Second_Phase_Text_Dataset
│   └── answer.txt
└── Validation_Dataset_Answer
    └── answer.txt

Training

We use bert-base-cased as the pretrained model.

python train.py

The model will be saved in the bert_finetuned_ner folder.

bert_finetuned_ner
├── checkpoint_12873
├── checkpoint_4291
└── checkpoint_8583

You can also download the pretrained model from here.

Please put the pretrained model in the bert_finetuned_ner folder.

Inference

python predict.py

The prediction will be temp.txt.