Research And Implementation Of Chatbot Based On Deep Learning

A chatbot is a program that gives an appropriate response based on a questioner's question in an open domain conversation.The richness of application scenarios and the effectiveness of information interaction will make chatbot applications more and more popular in future life.Therefore,the research on chatbot related technologies has great economic and social value.At present,chat robots still have problems of insufficient use of multimodal information,weak emotional expression ability,and lack of model fusion mechanism.With the wide application of deep learning in computer vision and natural language processing tasks,applying deep learning technology to chat robot application scene has become a new research point.The main research of this thesis is from the following perspectives:In view of the fact that the current chat robot products focus on the text level and lack the use of multi-modal information,the thesis proposes a multi-modal chat robot model based on Seq2Seq,which effectively uses multi-modal information such as text,picture and video.Compared with the single-modal model,the model has improved many indicators in the Weibo dataset experiment.Aiming at the problem that the current chat robot model pays more attention to the semantic fluency of generating sentences and neglecting the emotional expression ability,the thesis proposes a generative model that effectively uses user emotion information.By adding static emotion vector and emotion transfer model,Naturally outputting the appropriate response in the current emotional context,making the response more close to the person's emotional expression.Aiming at the single problem of the current chat robot product model,the thesis proposes a variety of mechanisms to integrate the search model and the generated model.On this basis,we apply the above scheme to the actual system to design and implement a complete,functional,efficient and practical chat robot system.