Prompt NER Chinese

Prompt-NER on Chinese EHR data

Introduction

This repository uses BERT as a pre-trained model to perform named entity recognition tasks using Prompt pre-trained method.

Folder description:

 .
├── .DS_Store
├── __init__.py
├── app.py 	# 接口文件
├── best_model.pth	# 模型文件（需要自己训练）
├── config.py	# 配置文件
├── conlleval.py	# 评价指标
├── data	# 数据集
│   ├── eval.txt # 处理好的验证集
│   ├── test.txt	# 处理好的测试集
│   └── train.txt	# 处理好的训练集
├── logger.py	# 日志文件
├── main.py	# 主文件
├── output	# 输出
│   └── logs
│       └── Experiment_log.log
├── predict.py	# 预测程序
├── processer.py	# 数据预处理文件
├── prompt_model.py	# 模型结构
├── test_predict.py	# 接口测试文件
└── utils.py	# 方法函数

Requirements:

python

torrh

sklearn

pandas

transformers

Direct usage method:

Run main.py with python to get the model file.

Run app.py with python and modify the input data in test.py to get the returned result.

result:

Training method	F1
Conventional pre-training methods	0.7617
Prompt training method	0.8189