Pre-trained Language Model based Medical Named Entity Recognition
Zheng Cheng, Jiaju Wu, Bin Ji, Huijun Liu · 2021 China Automation Congress (CAC) · 2021
Medical named entity recognition is an important part of structuring Chinese electronic medical records and construction of medical knowledge graph. The CCKS2019 conference organized a medical named entity recognition evaluation task to extract six types of medical entities from unstructured Chinese electronic medical records. Based on the data set of this evaluation task, pre-trained language model based entity recognition approaches are studied. First, select the BiLSTM-CRF model based on random initialized word embedding as the baseline system; secondly, apply word2vec to the baseline system; thirdly, apply ELMo to the baseline system. Experimental results show that the pre-trained language model is comparable to the best approach of this evaluation task, and the context-related pre-trained language model performs better.