融入领域信息的医疗领域命名实体识别研究

Named Entity Recognition Research on Medical Field Integrated with Domain Information

命名实体识别(Named Entity Recognition,NER)是自然语言处理(Natural Language Processing,NLP)中的一项基础工作,在医疗领域中,命名实体识别任务对病例信息抽取、医疗知识图谱的构建等工作有很大帮助。本文针对医学领域中一般的BERT方法不适合任务处理且标注数据太少的问题,采用了将领域信息融入字向量的方法,得到Domain-BERT-BiLSTM-CRF模型进行相关实验。结果表明对比使用原始BERT字向量的模型,新模型的F1值提高了2%。

Named Entity Recognition(NER)is a foundation work in Natural Language Processing(NLP).NER task can be of great help to extract case information and construct medical knowledge graph in the medical field.Aiming at the problems that the general BERT in the medical field is not fit for task processing and there are too few annotation data,this paper adopts the method of integrating domain information into word vector to obtain the Domain-BERT-BiLSTM-CRF model for relevant experiments.The results show that the F1 value of new model is increased by 2%compared with the original BERT word vector model.