Research And Application Of Emotional Dialogue Generation Algorithm Based On Seq2Seq

With the wave of artificial intelligence and big data,dialogue systems have become a research hotspot in the field of natural language processing and have attracted much attention from the industry.In the early days,template-based or retrieval methods were mainly used to construct dialogue systems,but such dialogue systems had disadvantages such as poor portability and insufficient intelligence.In recent years,with the rapid development of deep learning technology,dialogue systems have made breakthrough developments that can respond autonomously,but such responses lack emotional factors.Therefore,the realization of autonomous responses dependent on emotional factors is a current research hotspot in the field of dialog systems.At present,Seq2Seq(Sequence to Sequence)is the mainstream dialogue generation model in the field of natural language processing.This model is a breakthrough in the field of generative dialogue and can automatically generate responses.But there are also the following shortcomings: lack of contextual information,resulting in a mismatch between the reply content and the context;easy to produce safe replies,resulting in a relatively single model answer;the model lacks emotional factors,that is,the model does not consider the emotional connection between the question and the reply.Therefore,in response to the above problems,in order to realize the emotional interaction of the dialogue system,the specific work of this topic is as follows:(1)Aiming at the problems of the replies generated by the Seq2 Seq dialogue model,such as the mismatch of the context and the single reply,the attention mechanism and the search algorithm are improved to improve the diversity and relevance of the replies.(2)Aiming at the problem that the Seq2 Seq dialogue model cannot generate emotional response,this topic proposes an emotional assistance model,which is integrated with Seq2 Seq into a new emotional dialogue generation model to realize emotional interaction.The model can generate six emotional responses that are consistent with the emotion of the source sentence and specified {happy,like,sad,disgusted,angry,and others}.(3)The project combines automatic evaluation and manual evaluation to compare and analyze the language ability,generated response quality and accuracy of the model.The content response quality and emotional response quality of the improved emotional dialogue generation model are compared with the baseline model.Increased by 0.22 and 0.26.Experimental results show that the model can generate emotional and context-related responses.(4)Based on Tensorflow Serving and Docker deployment,an emotional dialogue generation system for family escort scenes was implemented,and multiple interactive modes of voice and Web were designed to verify the feasibility and effectiveness of the model in practical applications.Tests show that this emotional dialogue generation system can display dialogue content,communicate with users,and improve user satisfaction.