TRBNER: named entity recognition of TCM medical records based on multi-feature fusion

In the domain of natural language processing (NLP), named entity recognition (NER) within traditional Chinese medicine (TCM) medical records represents a significant challenge. TCM medical records, as a vital form of medical literature, currently encounter numerous obstacles, including the processing of unstructured data and the standardization of terminology. These issues impede the integration, retrieval, and application of knowledge contained within medical records, thereby constraining the advancement of TCM informatization. To address these challenges, we propose a multi-feature embedded named entity recognition model, referred to as TRBNER, specifically designed for TCM medical records. This model is based on the ALBERT architecture and introduces an enhanced Top-k attention score screening method. It integrates the first stroke features of Chinese characters, which are extracted using a convolutional neural network (CNN), with word boundary features for effective feature concatenation. Subsequently, a multi-scale channel attention module (MS-CAM) is employed for feature fusion, and the resultant features are processed through a Bidirectional Long Short-Term Memory (BiLSTM) network. The output is then refined by a conditional random field (CRF) layer to derive the optimal label sequence. Experimental results indicate that the TRBNER model demonstrates strong performance in the task of named entity recognition for Chinese medicine, achieving accuracy, recall, and F1 scores of 0.8844, 0.8675, and 0.8759, respectively, particularly when handling complex TCM texts.