Research On Key Technologies Of Intelligent Question Answering Oriented To Self-service Of Passengers In Civil Aviation Airport

Rapidly development of civil aviation in China,the passenger flow has increased sharply,and the competition between civil aviation is changing to service competition.An effort is made to build a Question Answering System of Civil Passenger Self-Service(QACAPS)to provide more convenient and faster online question and answer interaction for users.On the one hand,it improves the passenger service experience,on the other hand,it also reduces the cost of human and material resources of the airport.The main research work of this paper is as follows:(1)The construction of civil aviation passenger question answering knowledge base pro-vides data basis for question-and-answer system.The knowledge base consists of two parts:one part is named question knowledge base(QKB)with tens of thousands of common questions,other part is named answer knowledge base(AKB)with more than three hundred of standard answers.Then QKB is extracted by artificial recognition:firstly,a set of key phrases is extracted to represent the latent semantics of users for each question;secondly,the best answer is matched from AKB for each question.The AKB is divided into eight categories and cover most of the questions and answers.The data in QKB and AKB are transformed into a unified expression through pre-processing operations such as labeling category phrases,word segmentation,noise filtering and standardized translation to prepare for subsequent keyword extraction and question and answer matching.(2)Keyword extraction method based on TextRankSTM algorithm for civil aviation corpus provides a guarantee for the reliability of problem analysis.A word segmentation state transition matrix(STM)is obtained to express the civil aviation passenger's question and answer habits through model training.Combining STM and TextRank algorithm,TextRankSTM algorithm is designed to extract keywords.By comparing with the keywords labeled manually,the experiment proves that TextRankSTM algorithm has higher accuracy and recall rate than TextRank algorithm in the field of civil aviation passenger question answering,but TextRankSTM algorithm needs to invest relatively more human and material resources to label data manually.(3)A question and answer matching method is designed based on ChannelVSM model,and the optimal answer is retrieved and returned to the user.TextRankSTM algorithm and Vector Space Model(VSM)form the core of ChannelVSM model.Firstly,the questions and answers are processed into a set of keywords and transformed into sentence vectors.Then,the question cate-gories are identified by REC clustering method and the related answer list is returned.Finally,the best question-answer matching is determined according to the distance between the question vectors and each answer vector.Experiments show that the accuracy of REC clustering method is high,and Channel VSM Question Answer Matching Model achieves good results in practical application.