Research On Causality Of Aviation Safety Accident Based On Event Evolutionary Graph

With the development of the civil aviation industry,the reasons of aviation safety accidents are diversified and complicated,and the analysis of accident causality is facing new problems.The current research on the causality of aviation safety accidents lacks the analysis of the process of accident details and causal evolution.The event evolutionary graph is a kind of Logic Knowledge Base,that reveals the evolutionary laws and patterns between events,is applied to the causality research of aviation safety accidents.Therefore,this paper characterized the evolution model of aviation safety accidents by constructing event evolutionary graph,including:The method of extracting the causality of aviation safety accidents was studied on the tracking report of the World Civil Aviation Accident Investigation.According to whether the text contains causal prompts,the causal relationship can be divided into explicit causal relationship and implicit causal relationship.The pattern matching method was applied to get explicit causality.Experiments show that the pattern matching method can effectively extract the explicit causal relationship of aviation safety accidents,and the extraction accuracy rate reaches 87.72%.While the improved bidirectional long short-term memory networks with self-attention mechanism(selfAtt_BiLSTM),only the sentence set with the event trigger word position identifier is taken as input,was adopted for implicit causality extraction.This method is an improved for the method Att_BiLSTM,proposed by Zhou,taking advantage of the selfattention mechanism can fully capture more comprehensively semantic information of sentences.Experiments on the public data set and the domain data set show that the F value of the causality extraction is increased by nearly 6% and 10%,respectively.On the basis of causality extraction,the generation of the event evolutionary graph of aviation safety is realized.According to the results of the causality extraction of aviation safety accidents,the different expressions of the same event are combined by the event co-reference resolution,and then the causality graph for 955 accidents was generated,which can be used to describe the development process of the accident and reproduce the accident.Then the specific events described in the single accident causality diagram are abstracted,mapping with the domain ontology,to construct the aviation safety event evolutionary graph.The aviation safety event evolutionary graph can be used to describe the general laws and patterns of accident development,which support for relationship analysis and forecasting decisions.