| Machine reading comprehension is a technology to accurately acquire knowledge from a mass of fragmented information streams,and knowledge graph is a knowledge representation tool that contains entities,concepts and their semantic relationships.Its rich semantic relationships can lay the foundation for the management and application of domain knowledge.Therefore,it is of great significance to study the construction of knowledge graph of civil aviation emergency.Based on the investigation reports of civil aircraft incidents over the years in the accident database of China Civil Aviation Safety Information Network,the following studies focus on the knowledge acquisition of civil aviation emergency knowledge graph.Firstly,on the basis of in-depth analysis of machine reading comprehension methods,aiming at the problems that the semantic fusion of the passages and the problems in the current machine reading comprehension model is insufficient and the global semantic information is not fully considered,a machine reading comprehension model named BERT＿Att based on BERT,attention mechanism and self-attention mechanism is proposed.In this model,BERT is used to map passages and problems into feature space,and Bi-LSTM,attention mechanism and self-attention mechanism are used to fully integrate the semantic of passages and problems.Experimental results on the public dataset Du Reader show that the BLEU-4 and Rouge-L values of the model are improved by 3.2% and 2.1%,respectively,which provides a method support for semantic data acquisition of domain knowledge graph.Secondly,based on the thorough analysis of the content of the knowledge graph of civil aviation emergency,and aiming at the acquisition of complex knowledge,on the basis of BERT＿Att model,the domain knowledge acquisition model named KB＿ALBERT＿Att is constructed through the process of domain data preprocessing,improved vectorization of knowledge graph and data mapping.Experimental results on domain datasets show that the model can mine more complex instance data,and the overall EM value and F1 value reach88.5% and 91.4%,respectively.Therefore,it provides a more effective method support for the construction of high-quality knowledge graph of civil aviation emergency. |
