基于BERT的中文命名实体识别研究

Named Entity Recognition (NER) is an important research direction in the field of natural language processing (NLP). Chinese grammar is complex and word order is variable. The current mainstream model for named entity recognition is BiLSTM + CRF. This model still has the problem of poor generalization ability for more open Chinese scenes such as novel text. This article analyzes the characteristics of Chinese and named entity recognition tasks, and explores how to make the current popular research of NLP transfer learning and its representative work BERT better adapt to Chinese named entity recognition tasks, and proposes a BERT-based named entity recognition model BB- CRF. BB-CRF has the following advantages: strong generalization ability, can adapt to complex syntax and OOV problems, and fast convergence. Finally, this paper evaluates the model on two datasets, the People\'s Daily and the novel data closer to natural grammar. The experimental results prove that the model can model the characteristics of the Chinese named entity recognition task in a wide range of scenarios. , With high practicality and guiding significance.