Research On The Construction Method Of Event Evolution Graph Of Urban Rail Transit Operation Emergency

The emergency case of urban rail transit operation is an unstructured historical data described in text,which describes the occurrence and development process of urban rail transit operation emergencies,which contains the constituent elements of the event and information about the state of the event.Utilize the event information automatically extracted from the urban rail transit operation emergency case and the development process information of the event to construct a relationship diagram describing the evolution law and development logic of urban rail transit operation emergencies,which can restore the urban rail transit in a structured manner the evolution process of operational emergencies can discover the characteristics of the evolution process of urban rail transit operating emergencies,so as to better understand and grasp the development trend of the incident;at the same time,it provides the support of event evolution information data for experts and scholars in the field of emergency management.In order to make emergency decision-making activities correctly and quickly in the face of emergencies in urban rail transit operations,and strengthen the safety of urban rail transit operations.For this reason,this paper mainly studies the construction of event evolution graph of urban rail transit operation emergencies.Regarding the urban rail transit operation emergency cases obtained from the news webpage,firstly,the conceptual system of urban rail transit operation emergencies was defined,including event definition,event type definition and event relationship definition,etc.,and based on this,corpus labeling was carried out.Then use natural language processing technology to complete the automatic extraction of event information and event relations,and then use different visualization databases to store the obtained data and generate the urban rail transit operation emergency event evolution graph,which can be visually displayed.The main work content is as follows:(1)In view of the lack of a dedicated event case library in the urban rail transit field,this paper will collect emergency cases of urban rail transit operations from news webpages,perform data deduplication,denoising and data fusion,and construct urban rail transit operations emergency cases corpus.And according to the description characteristics of urban rail transit operation emergency cases,combined with expert knowledge,defined the concept system related to urban rail transit operation emergencies,and defined the natural disaster category,social safety category,accident disaster category,construction class,other accident category 5 kinds of event types;at the same time,this paper also puts forward the concept of "non-entity event element",defines the location,time,vehicle status,line condition,site condition,traffic impact,casualties,personnel detention,property damage 9 kinds of physical and non-entity event element roles for each event type;and 4 event relationship types are defined,including transition,condition,causality and succession relationship.Finally,based on the defined conceptual system,the emergency cases of urban rail transit operation are marked.(2)Aiming at the problem of high event density and scattered event elements in the urban rail transit operation emergency case description,combined with the event types and event element roles defined in the urban rail transit operation emergency related conceptual system,a document-level urban rail transit operation emergency extraction model is constructed,and the information of urban rail transit operation emergency and the elemental information contained in the event is studied from the whole case description.The model as a whole is regarded as a sequence labeling model.After the document vectorization of the labeled urban rail transit operation emergency case text,the convolutional neural network(CNN)is used to extract the word-level features of the sequence text that integrates contextual semantic information,and then through the Bi-directional Long Short-Term Memory network(Bi LSTM)obtains the contextual features of the sequence text,and then introduces the attention mechanism(Attention)to highlight the key semantic information features of the sequence text,completes the recognition of the trigger word of the urban rail transit operation emergency,obtains the event type,then the text is labeled for urban rail transit operation emergency cases that incorporate the characteristics of the identified trigger words,the context semantic information of the sequence text is obtained by using the attention mechanism-based Bi-directional Long Short-Term Memory network(Bi LSTM-Attention),and then the sequence label results are obtained with the conditional random field(CRF),so as to realize the identification of the elements of urban rail transit operation emergency and complete the extraction of urban rail transit operation emergencies.The experimental results show that compared with the proposed baseline method,the F1 value is increased by 4.65%(trigger word recognition)and 11.76%(event element recognition),respectively.(3)According to the event relationship types defined in the conceptual system of urban rail transit operation emergencies,as well as the method of dividing relational connectives,construct a triple representation of events,event elements,and events and event elements.The group has the problem of element overlap,so an urban rail transit operation emergency relationship extraction model based on the pre-training model BERT is proposed.The pre-training model BERT is used to learn the contextual semantic information of the event case annotation text,and then identify the first element in the event case.Finally,all possible related relationships of the first element and the corresponding tail elements of the relationship are identified,so as to realize the extraction of the relationship of emergency events in urban rail transit operations.The experimental results show that the F1 value is increased by 13.29% compared with the proposed baseline method.(4)Use Neo4 j graph database and protégé ontology development tools to store event information,event element information and relationship information obtained through urban rail transit operation emergency extraction and relationship extraction,and carry on the visual display of the event evolution graph of urban rail transit operation emergencies.Figure 30,table 19,reference 71.