方法对比

	基于领域自适应BERT的分类	域自适应 Transformer
领域	深度学习	深度学习
方法族	Machine learning	Machine learning
起源年份	2019–2020	2019–2022
提出者	Gururangan et al. (2020); earlier domain-specific instances include Lee et al. (2020) — BioBERT	Various (Vaswani et al. 2017 for Transformers; domain adaptation extensions emerged 2019–2022)
类型	Domain-adaptive pre-training followed by supervised fine-tuning	Pre-trained model fine-tuned with domain-shift adaptation
开创性文献	Gururangan, S., Marasovic, A., Swayamdipta, S., Lo, K., Beltagy, I., Downey, D., & Smith, N. A. (2020). Don't Stop Pretraining: Adapt Language Models to Domains and Tasks. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL 2020), 8342–8360.	Ni, J., Hernandez Abrego, G., Constant, N., Ma, J., Hall, K., Cer, D., & Yang, Y. (2021). Sentence-T5: Scalable Sentence Encoders from Pre-trained Text-to-Text Models. Findings of ACL 2022. arXiv:2108.08877.
别名	DAPT BERT classification, domain-adaptive pre-training, domain-specific BERT fine-tuning, BERT DAPT	DAT, domain-adaptive Transformer, domain adaptation with Transformers, transfer-learning Transformer
相关	6	2
摘要	Domain-adaptive BERT-based classification extends the standard fine-tuning pipeline by first continuing BERT's masked-language-model pre-training on a large corpus of in-domain unlabeled text, then fine-tuning the adapted model on labeled examples for the target classification task. This two-stage approach closes the vocabulary and distributional gap between BERT's general pre-training corpus and specialized domains such as biomedicine, law, finance, or social-media text.	A Domain-Adaptive Transformer (DAT) is a Transformer-based model — such as BERT or ViT — extended with an explicit domain-alignment objective so that learned representations transfer well from a labeled source domain to a different, often unlabeled, target domain. The approach combines the powerful representation capacity of Transformers with domain adaptation techniques such as adversarial training or contrastive alignment to minimise domain shift.