Research And Implementation Of Intelligent Dialogue Technology Based On Context Semantics

The development and wide application of deep learning technology has greatly promoted the development of natural language processing.the human-machine dialogue system,as a major application in the field of natural language processing,is a hot research topic in academia and industry today.With the development of applications such as intelligent voice assistants and navigation robots,human-machine dialogue will also become the next generation of the main way of human-computer interaction.This thesis takes the generative dialogue model as the research object,researches and applies the key technologies in the multi-turn dialogue generation.Research in recent years has found that multi-turn dialogue systems focus little on semantic information in historical contexts,leading to problems such as universal responses and poor quality.In this thesis,by introducing Self-Attention mechanism and external knowledge,the information of historical context is effectively obtained,and the quality of model response is improved.The specific research content of this thesis is as follows:(1)A generative dialogue model based on context information filtering is proposed.Considering the problem that common models do not use or distinguish historical context information,the context information filter based on Self-Attention mechanism is introduced to filter historical conversations at word level and sentence level,and different weights are assigned respectively.By the way proposed above,we grasp the useful information in the historical context from two granularity,enhance the influence of useful information on the model generation process,so as to achieve the purpose of improving the quality of model generation.Finally,a comparative experiment confirms that the replies generated by the model are more correlated with historical conversations,which improves the quality.(2)A two-step decoding generative dialogue model incorporating external knowledge is proposed.Based on the generative dialogue model based on context information filtering,the model is improved in order to further improve the quality of response generation.It is mainly divided into two aspects: one is to select the knowledge information related to the historical conversation from the knowledge base through the attention of the historical context;the other is to integrate the selected knowledge into the response generation through a two-step decoder.Finally,a comparative experiment proves that the model effectively integrates knowledge information,and the fusion effect of two-step decoding is better than the one-step decoding process of integrating information in advance,there by further improving the quality.(3)Based on the above work,finally designed and implemented an intelligent dialogue system for users' daily chat.Tests show that the model proposed in this paper has practical effects on the accuracy and richness of the generated responses,laying a foundation for the next step of research.