A Distribution-Free Framework for Rewrite-Based Human-text Detection via Knockoff Filtering

1. 1Formulate the text detection problem as a multiple hypothesis testing problem with knockoff structure
2. 2Convert arbitrary rewrite-based detectors into detectors with finite-sample FDR guarantees
3. 3Apply knockoff filtering to separate the design of detection statistics from the choice of detector
4. 4Test the framework using LLM-generated text and human-text samples
5. 5Evaluate the performance of the framework using metrics such as FDR and accuracy

Data scientists and AI engineers can benefit from this framework to improve the accuracy of text detection models, while researchers can use it to develop new detection statistics

💡 Knockoff filtering enables the conversion of arbitrary rewrite-based detectors into detectors with finite-sample FDR guarantees without retraining

🚀 Detect human-text from LLM-generated text with a distribution-free framework! 📊