Research On Text Classification And Reason Extraction Of Railway Traffic Accidents

As an important national infrastructure and key livelihood project,railway has effectively promoted the social and economic development.Safety of railway operation is an important safeguard of railway transportation.With the continuous improvement of railway transportation volume in China,attention of railway traffic accidents is also increasing.The effective identification of railway traffic accidents can provide a better guarantee for the post-processing of railway accidents and the analysis of railway accident safety.At present,most of the railway traffic information is based on text.It is difficult to distinguish the accident level and the cause of railway accidents.It is a waste of labor and time to judge the text of railway traffic accidents through the professional knowledge of railway professionals,and it is difficult to use it again in the following arrangement.There are three problems in the text processing of railway traffic accidents.First,because the professional in the field of railway,contains a lot of professional words,need railway professional judgment,defining the level of railway traffic accidents based on the accident text,with more professional word,describes the characteristics of the text length is differ,conventional text segmentation phase thesaurus will text errors of railway accidents or redundant word segmentation;Secondly,the single word vector text representation method has the strategy of "more give up and less complement",which makes the text sequence lack information in the neural network input,and the single sentence vector text representation method has less information.Finally,in the extraction of the causes of railway accidents,the output constraints of the single long-short time memory network are less,and the output sequence is not logical.The neural network intelligent classifier based on short and long time memory network can extract the complex features from the text by using the nonlinear activation function,which provides a new research direction for the railway traffic accident text.Long-short time memory networks relate long relationships in text according to their own gated neural network units.In response to the above questions,this paper takes the long-term and short-term memory network as the main network to carry out experiments on the text classification of railway traffic accidents and the text reason extraction of railway traffic accidents.The main research contents are as follows:(1)According to the professional knowledge of railway field and the difference between railway text thesaurus and conventional thesaurus,"railway traffic accident Thesaurus" is made.Considering the different directions of railway field,the constructed thesaurus contains five secondary word libraries,which improves the accuracy of Chinese text segmentation and neural network processing.(2)The network input method based on word vector text representation needs fixed network input data,which makes it lose part of text information when processing long text and when processing short text,so this paper designs a short and short time memory network structure combining word vector channel and sentence vector channel to capture text information more efficiently.In addition,static attention mechanism is introduced into word vector channel to further improve the grading accuracy.The experimental verification is carried out in the text of railway traffic accident with different length characteristics,and the feasibility of the method is proved by the comprehensive analysis of the experimental results of different algorithms.(3)Aiming at the problem that the extraction of railway accident causes is not clear,this paper marks the causes of four categories of accident texts: locomotive,electrical,public works and power supply.Through the combination of long-term and short-term memory network and conditional random position,the causes of railway accident texts are extracted.Conditional random position can make the output sequence according to logic,and improve the accuracy of railway accident causes extraction.