Medical Named Entity Recognition Based on Multi Feature Fusion of BERT

In order to solve the problem that traditional word vectors are difficult to express the context semantics and the feature extraction of traditional model is single, a multi-feature fusion model named BERT-BiLSTM-IDCNN-Attention-CRF for Named Entity Recognition is proposed, which uses BERT to model the context semantic relationship of word vectors and fuse the context features and local features extracted by BiLSTM and IDCNN respectively. The proposed model is tested on Chinese Electronic Medical Record (EMR) dataset issued by China Conference on Knowledge Graph and Semantic Computing 2020 (CCKS2020).Compared with the baseline models such as BiLSTM-CRF, the experiment on CCKS2020 data shows that BERT-BiLSTM-IDCNN-Attention-CRF achieves 1.27% improvement in F1. The experimental results show that the proposed model can better identify the medical entities in EMR.