Entity Recognition of Chinese Medical Literature Based on BiLSTM-CRF and Fusion Features
Jinshan Zhang, Yutai Wang, Xinghai Yang, Fengjiao Wang · 2020
The entity recognition status of Chinese medical literature is introduced in the paper. On the foundation, we propose a method based on fusion-feature to solve problems of medical named entity recognition. Then, the BiLSTM-CRF model is constructed by combining the Conditional Random Field (CRF) of Bidirectional Long Short-Term Memory (BiLSTM) network. It is used to obtain long-distance context information and label entity types. We use Word2vec to build the embedding layer. The characters and words in the sentence are converted into dense vectors with fusing external dictionary features. The results show that compared with using traditional CRF model, the method based on fusion-feature BiLSTM-CRF has better effect, and the F-measure is increased by 7.51%.