Research And Implementation Of Dialogue Generation Based On Controllable Emotion And Automatic Emotion

As one of the key branches in the field of artificial intelligence,manmachine dialogue has attracted the attention of academia and industry because of its huge imagination space and attractive commercial value.People have also explored this technology for a long time.The early manmachine dialogue is mostly based on finite state automata or template,which can communicate with people in specific fields.In recent years,with the rapid growth of social data on the Internet and the development of deep learning technology,the open domain dialogue system based on deep learning has gradually become the hot spot of academic work,and has made a breakthrough in language understanding and dialogue generation.However,most of the existing dialogue systems focus on improving the relevance of content,and still face challenges in integrating emotional information.The existing research on the emotional dialogue generation is mainly divided into two directions.One needs to take an emotion label as the extra input,so that the machine can generate sentences containing the specific emotion.The other needs to automatically perceive emotion in the process of multi-round dialogue and produce appropriate emotional response like human beings.The ultimate goal is to reproduce the habitual social behavior in human dialogue.This paper studies these two directions respectively,and the main works are as follows:1)A controllable emotional dialogue generation model based on part of speech analysis(PosEDG)is proposed.From the text of daily dialogue,we can find that the expression of emotion has obvious correlation with the part of speech of words.For example,adverbs and adjectives usually play more roles in expressing emotion in a sentence,while nouns and verbs play more roles in expressing content.Therefore,PosEDG uses a pretrained part of speech extractor to analyze part of speech features in the process of generating sentences,and uses a dynamic selector to dynamically embed emotion and semantics according to the part of speech features.The results of comparative experiment and ablation experiment show that PosEDG model has achieved better results in emotional accuracy,contextual relevance,dialogue fluency and other criteria.2)An automatic emotional dialogue generation model based on retrieval database(ReEDG)is proposed.In the encoding process,the emotion encoder not only relies on the dialogue that has occurred,but also introduces additional emotional information that can be referenced through the retrieval process.At the same time,it uses the emotional interactive attention mechanism to stimulate more appropriate emotional signals.The experimental results show that ReEDG can effectively track the dialogue context and produce more appropriate emotional and content responses.In addition,it performs well in grammar.3)Based on the above two models,an intelligent emotional dialogue system is designed and implemented.It can carry out emotional dialogue with users through two modes:specified emotion and automatic emotion,so as to provide spiritual companionship for users.Firstly,this thesis introduces the related work of emotional dialogue generation,and analyzes the problems in the existing methods.Then,it introduces the relevant theories and technologies needed to realize the algorithm in this paper.On this basis,it expounds in detail the controllable emotional dialogue generation model based on part of speech analysis and the automatic emotional dialogue generation model based on retrieval database.Based on the two models,an intelligent emotional dialogue system is designed and implemented.Finally,the work of this paper is summarized and the future research direction is prospected.