Chinese Medical Named Entity Recognition Enhanced with Glyphic Features

Medical entity recognition aims to identify medical entities in medical text data, including diseases, symptoms, medications, and physical examinations. Medical entity recognition can support knowledge graphs, question answering systems, and other downstream tasks. However, compared to languages like English, entities in Chinese face challenges such as lengthy descriptions and nested entity types. With respect to other NERs, healthcare entity recognition is required to address a number of additional complexities, including the use of highly specialized vocabulary and abbreviations, the potential for words to have multiple meanings, and a variety of other challenges. To address these challenges, we propose a medical entity recognition method that fuses text semantic features and glyph features. The semantic representation vector and glyph feature vector are obtained by the LERT module and glyph embedding module, respectively. These two vectors are then fused and fed into the BiLSTM layer to extract text features using contextual information. Finally, a global pointer is used to mark the entity. The experimental results show that our model achieved an Fl value of 82.84% on the CCKS2019 dataset, demonstrating excellent performance compared to traditional methods. The ablation experiment validated the effectiveness of each module.