Research On End-to-end Task-oriented Dialogue System Based On Deep Learning

In recent years,task-oriented dialogue systems have been widely used in the industry,and smart devices can effectively carry one or more dialogue platforms.At the same time,the proposal of the metaverse will undoubtedly increase the usage scenarios and frequency of human-computer dialogue in the virtual world.However,the field of human-machine dialogue still faces many challenges,such as model optimization methods,insufficient data sets,and dialogue response accuracy problems.In the way of constructing the dialogue model,the pipeline method needs to design the functions of each module separately,which is more complicated to implement.As the amount of dialogue data increases,the end-to-end task-oriented dialogue system relies on the characteristics of neural networks,which can capture complete dialogue information through a large amount of data training.Through the research of end-to-end task-oriented dialogue system,this paper finds that the existing models do not use enough information in the dialogue encoding and decoding stages.Therefore,we conducted a large number of experiments on multiple existing public datasets,and proposed new methods and optimization schemes.The main contributions of this paper are:(1)Analyze the multi-level memory network and Mem2 Seq network structure,evaluate their use in end-to-end task-oriented dialogue models,optimize the use of data from different sources in the encoding and decoding stages,and fine-tune their underlying structures,and conduct experiments on the public datasets In Car,Cam Rest and DSCT2 datasets to verify and evaluate our model,and finally achieve better results.(2)Through the research on word vector technology,based on the pre-trained word vector of the Bert model,the internal correlation of the word vector is analyzed,and the principal component analysis algorithm is used for optimization.And the optimized word vector data is used in the end-to-end task-oriented dialogue system model,and the performance of the model is better improved.(3)Through the web development technology,the end-to-end task-oriented dialogue model is embedded in the dialogue platform,the overall architecture is designed,and the related technologies are used to control the data direction.Conduct a comprehensive test of the implemented dialogue platform to verify our design ideas and the function of the dialogue platform.Finally,this paper conducts a comprehensive research on the model research and implementation technology of the end-to-end task-oriented dialogue system,which has a good reference for the field of human-computer dialogue.