基于双通道双向长短时记忆网络的铁路行车事故文本分类  

Text Classification of Railway Traffic Accidents Based on Dual-channel Bidirectional Long Short Term Memory Network

作　　者：韩广 卜桐 王明明 郑海青 孙晓云 金龙 HAN Guang;BU Tong;WANG Mingming;ZHENG Haiqing;SUN Xiaoyun;JIN Long(School of Electrical and Electronic Engineering,Shijiazhuang Tiedao University,Shijiazhuang 050000,China;Equipment&Technology Center,National Railway Administration of the People’s Republic of China,Beijing 100844,China)

机构地区：[1]石家庄铁道大学电气与电子工程学院,河北石家庄050000 [2]国家铁路局装备技术中心,北京100844

出　　处：《铁道学报》2021年第9期71-79,共9页Journal of the China Railway Society

基　　金：国家自然科学基金(51674169);河北省自然科学基金(F2019210243);河北省高等学校科学技术研究项目(QN2019031,ZD2019140,ZD2018039)。

摘　　要：铁路行车事故等级基于事故文本进行界定,具有专业词多、描述文本长短不一的特点,给事故等级分类带来挑战。针对该问题,提出一种结合双通道双向长短时记忆网络和注意力机制的铁路行车事故文本等级分类方法。首先,构建铁路事故词库,并将铁路事故文本表示为词向量和句向量;其次,设计结合词向量通道和句向量通道的长短时记忆网络结构,更加高效地抓取文本信息;最后,在词向量通道中引入静态注意力机制,进一步提升分级精度。实验结果表明,该方法在查准率、查全率和综合衡量指标三方面均取得了良好的效果,证明了其有效性。The level of railway traffic accidents is defined based on the accident text,which has the characteristics of large number professional words and different lengths of description texts,and brings challenges to the classification of accident level.In view of the above problems,this paper proposed a text classification method for railway traffic accidents combining dual-channel Long Short Term Memory network and attention mechanism.First,the railway accident vocabulary was constructed to express the railway accident text as word vectors and sentence vectors.Second,a LSTM network structure that combines word vector channel and sentence vector channel was designed to capture text information more efficiently.Finally,the static attention mechanism was introduced into the word vector channel to further improve the classification accuracy.The experimental results show good results achieved by the proposed method in three aspects:Precision,Recall and comprehensive measurement indicators F1,indicating the effectiveness of the method.

关 键 词：铁路行车事故 文本分类 长短时记忆网络 注意力机制 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]