Research On The Method Of Dialogue Generation For Traffic Information Service

The development of urban traffic has brought a large number of human-computer interaction application needs,including traffic condition query,alarm processing,traffic route navigation,etc.,so the traditional human-computer interaction processing method has been unable to meet the actual application needs.The application of human-computer dialogue technology in the field of traffic information service gradually improves the efficiency of traffic information service.However,there are still the following problems in the application of human-computer dialogue technology in the field of traffic information service:first,the dialogue content of users is very oral,and users' dialogue intention cannot be effectively extracted;Second,when the user dialogue content does not show obvious dialogue intention,often appear irrelevant answer situation.In view of these problems,this paper applies text classification and text generation technology to the human-computer dialogue system for traffic information service,identifying the user's conversational intent by text classification,and adjusting the conversation strategy according to the conversation intention category.At the same time,this paper adopts the text generation technology based on deep learning to reply to the dialogue contents with unclear intention,which improves the user experience and intelligence of the human-machine dialogue system for traffic information service and has great significance and application value.The specific research content is as follows:(1)A deep learning user dialogue intention recognition model is constructed.In this paper,the user intention recognition model based on the bidirectional long and short memory model and the convolutional neural network is constructed according to the dialogue text characteristics of short text and strong spoken in the field of traffic information service and the advantages of convolutional neural network and long and short memory model.Experiments show that this model can effectively extract the text features of user's dialogue content and improve the accuracy of user intention recognition.(2)A user intention recognition model based on transfer learning is proposed.In this paper,transfer learning is applied to the intention recognition in the field of traffic information service,and the user intention recognition model based on BERT is constructed,which solves the problem that the traffic information service field cannot meet the demand of deep learning model training due to the relatively small and difficult acquisition of certain application scene corpus.Experiments show that,in the case of a small amount of marked corpus data,the intention recognition model based on transfer learning constructed in this paper can achieve a relatively ideal effect of user intention recognition.(3)A dialogue generation model based on deep learning is constructed.By applying deep learning to generating dialog text,this paper constructs a BIGRU-based Seq2Seq text generation model,and improves the quality of conversation-generated content by introducing Attention,and solves the problem that when the traditional rule matching method is used to generate the dialog reply text and the user's conversation content intention is not clear,the question of answering the mismatch problem often occurs.(4)The human-computer dialogue experiment platform for traffic information service is built.This paper designs the dialogue strategy of the human-computer dialogue system for traffic information service,applies the above research results in this paper to the human-computer dialogue experiment platform,and evaluates the human-computer dialogue system by grading the text response content artificially.The experimental results show that the human-computer dialogue system performs well and can meet the practical application requirements.