Automatic Text Summarization And And Its Application In Aviation Safety Repors

Aviation safety report has a large amount of data,and involves all aspects of civil aviation production.Reading and understanding all these reports requires a wealth of professional knowledge and a long reading time,with a large investment in manpower and time.This paper attempts to study the automatic text summarization technology based on the deep learning method,and applies it to the automatic summarization generation of aviation safety report,which can save a lot of manpower to do the reading and writing work,reduce the working time to a certain extent,and solve the problem of high cost of manual summarization generation.This paper studies how to solve the problem of automatic text summarization based on generative adversary network and reinforcement learning,summarizes the common automatic text summarization models,and puts forward two generation methods.The main research work is as follows:1.In view of the problems of the supervised deep neural network method,such as error accumulation and low accuracy in generating aviation safety reports,this paper proposes a text automatic summary generation model based on the reinforcement of confrontation thought training.By using the generative adversary network discriminator as the reward function of reinforcement learning to guide the generation model optimization in time and avoid error accumulation as much as possible;by adding objective guidance features in the process of aviation safety report generation,help the abstract generation model acquire more text structure knowledge and improve the authenticity of the abstract generation model.The experimental results on the synthetic data and the aviation safety report data set show that the accuracy and authenticity of this method in the aviation safety report automatic summary generation task are further improved compared with the previous summary generation model.2.Aiming at the problem of information loss and low relevance between the abstract and the original text in the long text summary generation task of aviationsafety report,a text summary generation model based on cyclic generation countermeasure network is proposed for aviation safety report text summary generation task.Based on the objective guidance to ensure the rigorous text structure,the model uses the way of cyclic generation of confrontation network to ensure that the generated summary retains the original information of the report,and enhances the relevance between the generated summary and the original report,so as to solve the problem of information loss of the summary generation model of reinforcement of anti ideological training.The resulting Abstract combines the lossless information of the original text and ensures the strictness of the text structure.The experimental results on the task of short and long text summarization show that this method is more effective than the previous model in the task of automatic summarization of aviation safety report.