基于深度学习与词标签生成的病历文本后处理技术  

Medical record text post-processing technology based on deep learning and word label generation

作　　者：祁志玲 王晨宇[1] QI Zhi-ling;WANG Chen-yu(The First Affiliated Hospital of Hebei North University,Zhangjiakou 075000,Hebei Province,China)

机构地区：[1]河北北方学院附属第一医院,河北张家口075000

出　　处：《信息技术》2025年第3期56-60,共5页Information Technology

基　　金：河北省2022年度医学科学研究课题计划(20220589)。

摘　　要：为提高病历文本数据处理的效率,文中基于深度学习和词标签技术,设计了一种文本后处理技术算法。文中数据经过预处理后,被输入至卷积神经网络提取文本特征,通过构建医学词库,并根据专家评判从词库中挑选标签组成标准标签库。利用LDA主题模型获取词和标签的词-标签概率,采用双向长短时记忆网络得到词-标签概率特征,再将文本特征和词-标签概率特征进行特征拼接,使用全连接神经网络进行标签相似度筛选。与TextCNN和infoCNN模型相比,所提模型在Pr、Re和F1这三个指标上均有所提升,其中F1值分别提高了2.30%和2.22%,证明了该方案的可行性与优越性。To improve the efficiency of medical record text data processing,a text post-processing algorithm is designed based on deep learning and word tagging technology.After pre-processing,the text data is input into a convolutional neural network to extract text features,and a medical vocabulary is constructed.Based on expert evaluation,labels are selected from the vocabulary to form a standard label library.The LDA topic model is used to obtain word tag probabilities for words and labels,and a bidirectional long and short term memory network is adopted to obtain word tag probability features.Then,text features and word tag probability features are concatenating,and a fully connected neural network is used for label similarity screening.Compared with TextCNN and infoCNN models,the proposed model has improved in Pr,Re,and F1,with F1 values increased by 2.30%and 2.22%,respectively,proving the feasibility and superiority of this scheme.

关 键 词：病历文本 后数据处理 深度学习 卷积神经网络 LDA主题模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]