基于在线医疗平台的抑郁症严重程度预测  

Prediction of the depression severity based on online medical platforms

作　　者：张梓坪 林耿[2] ZHANG Ziping;LIN Geng(School of Computer and Information,Fujian Agriculture and Forestry University,Fuzhou 350002,China;School of Computer and Big Data,Minjiang University,Fuzhou 350108,China)

机构地区：[1]福建农林大学计算机与信息学院,福建福州350002 [2]闽江学院计算机与大数据学院,福建福州350108

出　　处：《河南工程学院学报(自然科学版)》2025年第1期65-72,共8页Journal of Henan University of Engineering:Natural Science Edition

基　　金：福建省自然科学基金项目(2024J011180;2023J05251)。

摘　　要：为解决现有抑郁症预测模型应对对抗性样本鲁棒性及类别标签敏感性不足的问题,提出了利用混合深度学习模型LAA-MBPCB预测抑郁症患者的严重程度。首先,使用医学中文双向编码器对文本数据进行编码,捕捉其上下文语义,并加入投影梯度下降对抗训练为数据添加扰动;然后,利用卷积神经网络和双向长短期记忆网络提取多粒度特征和长程依赖信息,用多层感知机提取年龄等非文本特征;最后,通过标签感知注意力机制优化特征权重。在某医疗平台患者问诊数据集上的训练与测试结果显示,LAA-MBPCB模型在准确率、召回率和F1分数等指标上均优于其他对比模型。该方法增强了模型对对抗性样本的鲁棒性和对类别标签的敏感性,为抑郁症严重程度预测提供了新的方法。In order to address the existing depression prediction model′s lack of robustness to adversarial samples and insufficient sensitivity to category labels,this study proposes a hybrid deep learning model,LAA-MBPCB,to predict the severity of depression in patients.Firstly,the medical Chinese bidirectional encoder representations from transformers(MC-BERT)is employed to encode the text data and capture its contextual semantics,while projected gradient descent(PGD)adversarial training is applied to introduce perturbations to the data.Secondly,convolutional neural networks(CNN)and bidirectional long short-term memory networks(BiLSTM)are utilized to extract multi-granular features and long-range dependency information,with multi-layer perceptron(MLP)used to extract non-text features such as age.Finally,the feature weights are optimized using a label-aware attention mechanism.Through training and testing on the patient consultation dataset of a medical platform,experimental results demonstrate that the LAA-MBPCB model outperforms other existing models in terms of accuracy,recall,and F 1-score.This method enhances the model′s robustness to adversarial samples and its sensitivity to category labels,offering a new approach for predicting the severity of depression.

关 键 词：在线医疗平台 抑郁症 对抗训练 标签感知注意力 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]