Electronic Medical Record Text Classification Based on Dual Attention Mechanism

With the digitization of medical information and the rapid growth of medical big data, electronic medical records (EMR) have become an indispensable part of the medical field. The secondary application of electronic medical records is focused on assisting medical diagnosis to improve the accuracy of clinical diagnosis. In this study, we propose a method based on dual attention mechanisms to identify key information in free-text electronic medical records. Firstly, we introduce a gated attention AG module to selectively adjust the attention weights of different positions in the input sequence, enabling the model to better focus on key information. Additionally, we incorporate the SE-Res module with residual units to enhance the representation capability of key text features and improve classification performance. We adopt a Transformer-based model and fine-tune it using the pre-trained BERT model to adapt to our classification task. Experimental results on the publicly available dataset of CCKS2018 electronic medical records demonstrate that the proposed method achieves an accuracy of 0.9812, recall of 0.9793, F1-score of 0.9836, confusion matrix, and average accuracy of 0.9811, effectively enhancing the effectiveness of text classification.