Contrastive Learning for Authorship Verification
A ModernBERT bi-encoder using contrastive learning reaches 98.4% accuracy on PAN21 authorship verification.
Researchers report that contrastive learning outperforms a classification baseline for authorship verification. They identify loss function, batch size, training duration, pretrained model, input context length, and random text-span augmentation as important performance factors. Their ModernBERT bi-encoder reaches 98.4% accuracy on the PAN21 authorship verification task.
- Contrastive learning outperformed classification for authorship verification in the tested settings.
- A ModernBERT bi-encoder reached 98.4 percent accuracy on PAN21.
- Loss, batch size, duration, model, context length, and span augmentation were key factors.
Full article62 words · extracted from arxiv.org · click to collapse
Our results show that contrastive learning outperforms a classification-based approach to authorship verification under the tested settings. We identify loss function, batch size, training duration, pre-trained model, input context length, and random text span data augmentation as important factors of model performance. Based on these considerations, we develop a ModernBERT Bi-Encoder model that achieves 98.4% accuracy on the PAN21 authorship verification task.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.28471