Medical Named Entity Recognition from Un-labelled Medical Records based on Pre-trained Language Models and Domain Dictionary

Medical named entity recognition (NER) is an area in which medical entities are recognized from texts, such as diseases, drugs, surgery reports, anatomical parts, and examination documents. Conventional NER methods do not make full use of un-labelled texts embedded To address this issue, we proposed a approach based on pre-trained language models domain dictionary. First, constructed dictionary by extracting labelled collecting other resources, the Yidu-N4K data set. Second, employed to train domain-specific using texts. Third, pseudo labelling mechanism automatically annotate create labels. Fourth, BiLSTM-CRF sequence tagging model was used fine-tune models. Our experiments were extracted Chinese electronic records, show that enables strict relaxed F1 scores be 88.7% 95.3%, respectively.