Research on Constructing a Departmental Guidance Model Using TextCNN, BiLSTM, and BERT

Current departmental guidance platforms in intelligent systems predominantly rely on selective options, necessitating patients to select pertinent medical information before scheduling an appointment. However, most patients' lack of specialized medical knowledge frequently results in misregistration incidents. Moreover, the focus of existing research on medical guidance systems does not emphasize or distinguish among specialized departments, leading to reduced consultation efficiency, overstretched resources in specific departments, and potential delays in patient treatment. This study aims to develop a departmental guidance model utilizing TextCNN, BiLSTM, and BERT. Initially, it involves training the base models of TextCNN, BiLSTM, and BERT on departmental guidance under uniform hyperparameters and environmental settings, comparing the outcomes to ascertain each model's strengths and weaknesses. Subsequently, an innovative hybrid model is proposed incorporating an attention mechanism, capitalizing on their synergistic features to address the challenge of low accuracy in guiding specialized departments. This hybrid model achieved an accuracy rate of 96.0%, precision of 95.7%, recall of 95.5%, and an F1 score of 95.6%, all demonstrating superior performance and significantly surpassing the experimental results of the standalone foundational models. Therefore, this study concludes that the innovative model can accurately direct patients to the appropriate specialized department based on their symptom descriptions.