Differentiate ChatGPT-generated and Human-written Medical Texts

Data Description

This paper involves two datasets: medical abstract (in data/medical_text) and radiology report (in data/MiMic) datasets.

all_ Data.csv contains all human-written data and ChatGPT-generated data

prompt*_seed*_train.csv, prompt*_seed*_val.csv, prompt*_seed*_test.csv is the training set, validation set, and testing set for different groups.

Software environment

pip install -r requirements.txt

run code

1. vocabulary and sentence analysis: python word_count.py
2. Part-of-speech analysis: python pos_analysis.py
3. Dependency parsing: python dependency_analysis.py
4. Sentiment analysis: python sentiment_analysis.py
5. Text perplexity: python PPL_distribution.py
6. Perplexity-CLS: python ppl_cls.py
7. CART: python CART_cls.py
8. XGBoost: python xgboost_cls.py
9. BERT: python BERT_cls.py'