STADEE: STAtistics-based DEEp Detection of Machine Generated Text
This addresses the need for more generalizable detection tools against machine-generated text, though it appears incremental as it builds on existing statistical and deep learning approaches.
The paper tackles the problem of detecting machine-generated text by proposing STADEE, a method that integrates statistical features with a deep classifier, achieving an 87.05% F1 score in-domain and outperforming existing methods in out-of-domain and in-the-wild scenarios.
We present STADEE, a \textbf{STA}tistics-based \textbf{DEE}p detection method to identify machine-generated text, addressing the limitations of current methods that rely heavily on fine-tuning pre-trained language models (PLMs). STADEE integrates key statistical text features with a deep classifier, focusing on aspects like token probability and cumulative probability, crucial for handling nucleus sampling. Tested across diverse datasets and scenarios (in-domain, out-of-domain, and in-the-wild), STADEE demonstrates superior performance, achieving an 87.05% F1 score in-domain and outperforming both traditional statistical methods and fine-tuned PLMs, especially in out-of-domain and in-the-wild settings, highlighting its effectiveness and generalizability.