基于ALBERT-LDA模型的康复训练方案推荐系统研究  

Research on rehabilitation training program recommendation system based on ALBERT-LDA model

作　　者：祝晓状 徐钱钱 高诺[1] ZHU Xiaozhuang;XU Qianqian;GAO Nuo(College of Information and Electrical Engineering,Shandong Jianzhu University,Jinan 250101,China)

机构地区：[1]山东建筑大学信息与电气工程学院,济南250101

出　　处：《生物医学工程研究》2024年第6期445-455,共11页Journal Of Biomedical Engineering Research

基　　金：山东省自然科学基金项目(ZR2022MF309);山东省科技型中小企业创新能力提升工程项目(2022TSGC2554)。

摘　　要：为解决传统康复训练方案因忽略了身体评价中上下文语义信息与潜在语义,导致语义表达不准确、训练方案推荐精确率较低的问题,本研究提出一种融合ALBERT和潜在狄利克雷分布(latent Dirichlet allocation,LDA)的深度学习网络模型(ALBERT-LDA),并基于该模型构建了肢体运动康复训练方案推荐系统。首先,该系统利用LDA主题模型和ALBERT模型分别获得文档级的主题信息和词级的语义表示;其次,采用TextCNN提取身体评价文本词级的语义特征,并通过分层注意力机制对提取的特征进行重建;最后,利用多重融合策略将重建特征与主题特征融合,推荐训练方案。实验表明,本研究提出的系统不仅能得到更深层次的语义特征,还能更全面地理解上下文语义,可为患者提供更客观、有效的康复训练方案。In order to solve the problem of inaccurate semantic expression and low recommendation accuracy rate of training scheme caused by ignore contextual semantic information and potential semantic in body evaluation in traditional rehabilitation trainig schemes,we proposed a deep learning network model(ALBERT-LDA)integrating ALBERT and latent Dirichlet allocation(LDA).Based on this model,a recommendation system for physical exercise rehabilitation training was constructed.Firstly,the system used the LDA topic model and ALBERT model to obtain the document-level topic information and the word-level semantic representation,respectively.Secondly,TextCNN was used to extract the semantic features of body evaluation text at word level,and the extracted features were reconstructed through hierarchical attention mechanism.Finally,a multiple fusion strategy was used to fuse the reconstructed features with the subject features to recommend training schemes.The experiment showes that the proposed system can not only obtain deeper semantic features,but also understand the contextual semantics more comprehensively,which can provide more objective and effective rehabilitation training programs for patients.

关 键 词：康复训练方案推荐 ALBERT模型 LDA主题模型 特征融合 

分 类 号：R318[医药卫生—生物医学工程] TP391[医药卫生—基础医学]