Design And Implementation Of Open Domain Dialogue System Integrating Conversational Emotion

With the rapid development of machine learning,deep learning technology,massive and complex dialogue data are stored in the network,which provides a strong support for man-machine dialogue system.Emotional dialogue system is a subdivision task in human-computer dialogue system,which aims to make the dialogue system have the ability to perceive and express emotions.The emotional chat system can enhance user satisfaction and contribute to positive perception interaction at the same time,which avoids misunderstandings due to emotional problems in the chat dialogue system.This system takes into account the characteristics of emotion and intelligence.In recent years,more efforts have been devoted to the research of large-scale dialogue generation methods based on deep learning,which are aimed at solving the quality of dialogue generated content.However,many of them fail to fully consider the emotional factors of dialogue generated content.Thus,there are still many problems.For example,in the process of dialogue model generation,the model can only generate emotional dialogue in a passive and low frequency way.Even if the generated dialogue contains emotion,it is usually the response generated by the effect of the dense emotional vocabulary data set.In order to solve many problems in the current emotional dialogue system,this paper studies the dialogue generation of specified emotion and non specified emotion based on deep learning technology on the data set of NLPCC and NLPCC2017 emotional dialogue generation tasks.The specific work of this paper is as follows :First,from the perspective of part of speech,this paper proposes an emotion supervision assisted generation model of designated emotion dialogue.The model consists of three parts:(1)Multi-emotion dialogue responder,which generate output sequences containing multiple emotions;(2)Emotional discriminator,which judges the emotional polarity hidden in the input sequence;(3)The gating mechanism selects the best output sequence based on the output results of the above two parts and uses it as the final output sequence.The experimental results show that it can not only produce appropriate output sequence in content(related grammar),but also produce appropriate emotional response in emotion(emotional consistency).Secondly,a non-specified emotional dialogue generation model(DESG)assisted by emotional supervision is proposed.On the basis of seq2 seq framework,DESG integrates dictionary based attention mechanism,which encourages the replacement of words in response with synonyms in emotion dictionary.At the same time,in order to improve the model,two mechanisms are introduced:(1)Internal emotion regulator :Using embedded emotion categories to express high-level abstract emotions,and capture changes in the internal implicit emotional state.(2)Sentiment classifier: Guides the response generation process to ensure that the generated output sequence contains the appropriate emotion type.At the same time,this paper not only uses a variety of sorting algorithm to improve the diversity of output sequence,but also uses an emotional dictionary to accurately express human experience and perception.The experimental results show that,for a given input sequence and emotion category,the model can not only produce an appropriate output sequence in terms of content(relevance grammar),but also express the expected emotional response explicitly or implicitly in terms of emotion.Thirdly,based on the above work,this paper designs and implements an open domain dialogue system integrating conversational emotion.The test results show that the model proposed in this paper has practical effect on the richness of conversation expression and the improvement of sentence generation quality,which lays the foundation for further research.