Research Article

Self-Information Loss Compensation Learning for Machine-Generated Text Detection

Table 4

Comparison of results with two initialization parameter methods.

Sequence length64128256512
Our modelF1 scoreACCF1 scoreACCF1 scoreACCF1 scoreACC

Random initialization parameters0.81470.82830.86330.86830.86400.86830.86590.8708
Pretrained parameters0.87310.87920.90720.91030.90030.90500.91310.9158