Domain-Adaptive Pre-training BERT Model for Test and Identification Domain NER Task

Name entity recognition (NER) is the foundation of a wide range natural language processing (NLP) task in domain test identification. In this paper, we continue to train pre-trained BERT model by unlabeled texts related identification, so as inject knowledge into and realize adaptation. The experiment results show that proposed domain-adaptive pre-training method increases F1 value 1% compared with baseline identification NER task.