基于深度学习的医疗文本分类模型  

Medical text classification model based on deep learning

作　　者：彭清泉 王丹[2] PENG Qingquan;WANG Dan(Department of Nephrology,Tongji Hospital,Tongji Medical College,Huazhong University of Science and Technology,Wuhan 430030,China;Departments of Geriatrics,Tongji Hospital,Tongji Medical College,Huazhong University of Science and Technology,Wuhan 430030,China)

机构地区：[1]华中科技大学同济医学院附属同济医院肾内科,湖北武汉430030 [2]华中科技大学同济医学院附属同济医院综合医疗科,湖北武汉430030

出　　处：《电子设计工程》2023年第5期51-54,60,共5页Electronic Design Engineering

基　　金：全国教育科学国防军事教育学科“十二五”规划教育部重点课题(DRA110425)。

摘　　要：针对静态词向量模型语义表示质量不高,深度学习模型无法聚焦关键特征等问题,提出了基于ChineseBERT-BiSRU-AT的医疗文本分类模型。预训练模型ChineseBERT融入字形和拼音特征,通过参考词的具体上下文语境,学习到词的动态语义表示。BiSRU模块提取文本高维序列特征,软注意力机制赋予关键词更高权重。在影像报告文本数据集进行实验,结果表明Chinese-BiSRU-AT模型取得了最高的F1分数,BiSRU模块训练效率更优,ChineseBERT模型应用效果更佳。To address the problems that the poor semantic representation of static word vector models,and the inability of deep learning model to focus on key features,a medical text classification model based on ChineseBERT-BiSRU-AT is proposed. The pre-training model ChineseBERT combines the characteristics of font and Pinyin,and learns the dynamic semantic representation of words through the specific context of reference words. BiSRU module extracts the high-dimensional features of text sequence,soft attention mechanism assigns higher weight to keywords. Experimental result on image report text datasets show that ChineseBERT-BiSRU-AT model achieves the highest F1 score,BiSRU module is more efficiency,and the application effect of ChineseBERT model is better.

关 键 词：文本分类 ChineseBERT BiSRU 软注意力 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]