This is the code of the article titled "C-Net: A Compression-based Lightweight Network for Machine-Generated Text Detection"

The overall framework of C-Net is shown in the figure

The portion of npz_gzip in this code is referenced inACL2023

Data preprocessing

Our raw data comes from HC3, which can be obtained from [https: //github.com/Hello-SimpleAI/chatgpt-comparison-detection.]()

To obtain the machine generated text for the corresponding problem, first run data_process to obtain the machine text field

python data_process.py --data_path chatgpt/unfilter_full/en_train.csv --save_path chatgpt/unfilter_full/train.csv --api_key your_openai_key

The processed training data should include at least three fields: label, answer, and machine_text

Train

python train.py

Test

Use the trained model from the above process for testing python Test.py

Citation

@ARTICLE{10508945,
  author={Zhou, Yinghan and Wen, Juan and Jia, Jianghao and Gao, Liting and Zhang, Ziwei},
  journal={IEEE Signal Processing Letters}, 
  title={C-Net: A Compression-Based Lightweight Network for Machine-Generated Text Detection}, 
  year={2024},
  volume={31}, 
  number={},
  pages={1269-1273},
  keywords={Training;Text detection;Feature extraction;Closed box;Task analysis;Loss measurement;Feeds;Compression;machine generated text detection;text classification},
  doi={10.1109/LSP.2024.3394264}}