Design And Implementation Of Event Graph Platform For The Field Of Emergency

The event graph is different from the existing knowledge graph.It is a graph with events as nodes and the relationship between events as the edge.According to the characteristics of the emergency,the combination of the emergency and the logic can better describe the evolution of the emergency.In addition,experts and scholars can quickly collect the relevant information of the emergency by using the event graph,and clarify the context of the evolution of the incident,so as to provide an efficient factual basis for the emergency decision-making plan,so as to control or mitigate the harm caused by the emergency.According to the characteristics of emergencies,this thesis divides them into three categories: event occurrence,event impact and event emergency decision,and designs and implements an event graph platform for emergencies.This thesis mainly studies the related technologies in the process of constructing the event graph.The specific contents are as follows:(1)A method of joint identification of entities and emergency trigger words based on sequence annotation is proposed.The sequential annotation model is used to identify the entity and trigger words of the text sentence.The experiment shows that this method can improve the effect of emergency identification.(2)Different event element extraction methods are adopted for different types of emergencies.For event occurrence class,this thesis proposes an event element extraction method that combines the characteristics of entity and event trigger words,and uses the dependency relationship between entity,event trigger words and event elements to extract event elements.The experiment shows that this method improves the effect of event element extraction.For event impact class and event emergency decision class,this thesis proposes an event element extraction method based on dependency syntactic analysis,which determines the core composition in the sentence through dependency syntactic analysis,and according to certain constraint rules,extracts the core vocabulary containing such event trigger words.This method can effectively extract the event elements.(3)A causality extraction method of emergencies based on pattern matching and neural network is proposed.Firstly,two groups will be drawn for emergency events.Then,in the case that two emergency events point to the same event description sentence and there are causal trigger words,the extraction method of pattern matching is used to extract causality.Otherwise,it is regarded as a classification problem and the causality of emergencies is classified by using neural network.Finally,the results are extracted into the form of a pair of causal events.The experiment shows that this method can improve the effect of causality extraction.(4)An emergency fusion method based on event synthetic similarity is proposed.According to the characteristics of emergency,the comprehensive similarity of emergency is calculated from the similarity of event description sentences and the cooccurrence of entities for the event impact class and event emergency decision class.For the event occurrence class,the comprehensive similarity of events is calculated from three aspects: the similarity of event descriptors,the similarity of event elements and the cooccurrence of event descriptors.Finally,the synthetic similarity of events is used for the fusion of emergencies.The experiment shows that this method can effectively fuse events.(5)A event graph platform for the field of emergency is designed and implemented,and the function of the platform was tested.In this thesis,the graph platform is divided into three parts: data collection platform,data annotation platform and graph construction platform.The data collection platform provides real acquisition data for the graph construction platform,and the annotation platform provides annotation data for the graph construction platform.The graph construction platform uses the annotation data and real data for graph construction and visualization.The cooperation of the three platforms constitutes a complete event map platform.