Civil Aviation Airport Based On Deep Learning Can Only Answer Key Technology Research

With the rapid development of civil aviation in China,there has been a significant increase in the number of civil aviation passengers,and the higher the demand of civil aviation passengers for aviation information.Traditional civil aviation airport customer service has been unable to meet the needs of passengers,how to more convenient and fast to provide passengers with information becomes particularly important.Intelligent question and answer system can not only help the airport reduce the cost input,but also improve the passenger service comfort,is the development trend of civil airport customer service.Therefore,the study of efficient intelligent question and answer in civil aviation airport customer service is of great significance.First,in view of the insufficient Q&A data of civil aviation,this thesispreliminarily obtains Q&A data of civil aviation airport through the investigation of several civil aviation information networks and the use of crawler technology.And through the mixed data enhancement,the question and answer data set of civil aviation is constructed to make data preparation for the experiment.Secondly,in view of the low accuracy of the traditional question answering model and the poor interpretability of the existing question answering model framework based on deep learning,a fusion of the deep learning model and the traditional information retrieval technology is proposed.By combining the advantages of the two technologies,a question answering model based on keyword extraction and answer sentence weight is proposed.Firstly,the vector representation of problem words in the problem is used,and then the weight of key words and the weight of answer sentences are obtained.Match the key words of the question and the weight of the answer sentence.Thirdly,in view of the poor parallel effect of the traditional neural network model and imperfect semantic information extraction in Chinese,this thesis uses Enhanced Representation through Knowledge Integration(ERNIE)as the model for semantic extraction.ERNIE can carry out parallel operation,and can integrate position information and context information with word vector,and also obtain Chinese phrase information,which can obtain more abundant word vector information.Fourthly,in view of the traditional information retrieval technology can not solve the problem of word gap and low retrieval accuracy,this thesis proposes an improved information retrieval technology based on BM25 F algorithm to achieve answer retrieval based on keywords and answer sentence weights.For the keywords extracted by neural network,the BM25 F algorithm based on the weight of answer sentences can solve the answer matching and make the answer matching more interpretable.