基于双向长短期记忆网络的医院电子病历数据挖掘

DATA MINING FOR ELECTRONIC MEDICAL RECORDS OF HOSPITAL BASED ON BIDIRECTIONAL LONG SHORT TERM MEMORY NETWORK

在当前信息化智慧医院系统中,临床病历和门诊病历是追踪诊疗进度的重要依据,挖掘电子病历数据有助于专家对病情的掌控。针对这种情况,基于长短期记忆网络提出电子病历方面级观点分析系统,智能地分析诊断过程的治疗效果。借助现有工具识别电子病历的中文命名实体,过滤与医疗领域相关的方面项;提取每个句子的三种语法特征,将句子和相关特征送入长短期记忆网络进行分析;结合注意力机制对句子的上下文语义进行分析,识别每个方面项的情感极性。在江苏省第二中医院的电子病历数据集上完成了验证实验,结果表明该算法在中文电子病历数据上实现了较好的性能。

In current informationization intelligent hospital systems,clinical records and outpatient medical records are the important reference for tracking the progress of diagnosis and treatment,and mining electronic medical records helps experts to control the patient s condition.In view of this,we propose an aspect-based opinion mining system of electronic medical records based on LSTM network to intelligently analyze the effects of diagnosis and treatment.The system recognized named entity of electronic medical records with the help of the existing tools,and it selected the aspect terms related to medical field.The algorithm extracted three syntactic features of each sentence,and delivered each sentence with corresponding features to LSTM network to analyze.An attention mechanism was combined to analyze the semantic context for each sentence to recognize the emotional orientation for each aspect.Validation experiments were carried out on the electronic medical records dataset of Jiangsu Provincial Second Chinese Medicine Hospital.The results indicate that the proposed algorithm realizes better mining performance on Chinese electronic medical records.