Research On Multi-turn Dialogue System Model Based On Transformer Frame

In terms of the problem that human language is typical of elliptical sentences and short sentences,which is difficult to express ideas accurately,the Multi-turn Dialogue System Model based on Transformer Frame(MDBOT)is proposed in this study.Based on the background of the application of task-oriented human-to-machine dialogue system in multiple fields,the prototype system of human-to-machine dialogue system is devised,which embeds the constructed models into the platform,aiming to explore the employment of this model in real life.To start with,the recognizing module is designed by using Term Frequency-Inverse Document Frequency(TF-IDF)algorithm and the Long Short-Term Memory(LSTM)is constructed to realize the design of the slot filling module.Then comes the intentional recognition of multi-turn dialogue.The cognition of the advantages of different algorithms in human-to-machine dialogue design is formed by using intentional recognition algorithm to realize the intentional recognition module design.And the intentional recognition algorithm is used to form the technical support for the multi-turn dialogue system construction.What's more,the analysis of the background of the multi-turn dialogue system model construction is unfolding.To introduce prior knowledge into transformer framework.The construction of MDBOT is accomplished by analyzing the frame of algorithm and encoding module,the frame of memory module,text content generation model and summary generation model.Combined with the training methods of employment of the dynamic word vector and the employment of the decoder to decode so that it can reach the aim of forming the target text abstract.At last,the analysis of the design of situational application to well-constructed models is processed so as to verify the feasibility and advantages of the model.In this study,ROUGE value is used as evaluation indices.The result of this study indicates that every index of models proposed in this study outnumbers that of benchmark model.The relevant information can be connected in series in the multi-turn dialogue system and fluent enough to accomplish the dialogue,which verifies the applicability of the multi-turn dialogue system.This thesis had 23 figures,8 tables and 60 references.