Named Entity Recognition for Instructions of Chinese Medicine Based on Pre-trained Language Model

• Published in: 2021 3rd International Conference on Natural Language Processing (ICNLP) pp. 139 - 144
• Main Authors: Wen, Song, Zeng, Bi, Liao, Wenxiong
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-03-2021
• Subjects: CRF; Data models; Drugs; Electronic medical records; instruction of Chinese medicine; Natural language processing; NER; pre-trained language model; Task analysis; Text recognition
• DOI: 10.1109/ICNLP52887.2021.00029

Named Entity Recognition (NER) of Chinese medicine text is a basic task of constructing medical and health knowledge graph. Many scholars have researched the NER task of electronic medical records and drug names, while many factors restrict the research of NER tasks for the instructions of Chinese medicine. For example, there is no obvious boundary between words in Chinese, and it is impossible to capture the interactive information between sentences and the global information at the same time. Considering that this type of data is highly professional and there is no publicly available data set. This paper collected 1,000 pieces of instructions of Chinese medicine, then explored the effectiveness of pre-trained models in NER task in this field. The experimental results showed that compared with the experimental results of the single or joint model on the same data set, the F1 value of pre-trained model was increased by 9.65% and 8.71% respectively.