Research And Implementation On End-to-End Task-Oriented Dialog Systems

The task-oriented dialogue system is an important research content in the field of natural language processing.In recent years,thanks to the increasingly mature deep learning technology,its related theories and technologies have made breakthrough progress.The endto-end approach to constructing a task-oriented dialogue system has brought a great surprise to the whole field of natural language processing.In order to follow up the research in this field and promote the development and application of this field,this paper focuses on the research and implementation of the end-to-end task-oriented dialogue system.This paper introduces and analyzes its basic theory,research status,and related technologies,and deeply analyzes the existing problems and challenges of the task-oriented dialogue system.The task-oriented dialogue model based on an end-to-end memory network is proposed in this paper and introduces the Glo Ve word vector to further improve the performance of the model.Finally,the proposed model is embedded into the dialogue platform,and the development platform of the task-oriented dialogue system is designed and implemented.This paper summarizes the main contributions as follows:(1)The previous model directly uses the output of the encoder as the query vector to retrieve external knowledge,which leads to the fusion of a lot of information unrelated to the retrieval of external knowledge,so it is difficult to integrate external knowledge effectively.To solve the problem of poor accuracy of entity retrieval in external knowledge,this paper designs a two-stage decoding strategy.In the first decoding stage,entity information is decoded from the context of the dialogue and used to guide the reading and writing of external knowledge;in the second decoding stage,dialogue responses are sequentially decoded based on the query results of external knowledge and the understanding of user intention.Experimental results show that the two-stage decoding strategy can better search external knowledge and generate high-quality responses.(2)The previous model uses Bi-GRU or Bi-LSTM to encode the dialogue context.This sequential structure of the encoder is easy to cause the loss of long dialogue information,so the model cannot accurately identify the user’s intention,which ultimately affects the accuracy of dialogue response generation.To solve this problem,an encoder based on the BERT structure is designed to capture important features in the dialogue context using its multi-head self-attention mechanism,which can enhance the semantic extraction ability of the task-based task-oriented dialogue model and enable the model to better understand the user intention.Through experimental verification and analysis,it is proved that the encoder based on the BERT structure has certain advantages.(3)Based on the proposed model,a task-oriented dialogue system is designed and implemented in this paper to meet the application requirements of man-machine dialogue in real scenarios.In the design process,fully consider user requirements,technology selection,architecture design,test deployment,and other aspects of the problem,to achieve a beautiful,usable,efficient,and stable task-oriented dialogue system.Through the performance and function test of the constructed task-oriented dialog system,it is verified that the system can meet the basic functional requirements of users and that the dialog experience is good.