Research On Emotion Recognition Technology For Chatbot

In recent years,with the rapid development of natural language processing technology,chatbot has received much attention,and has gradually become a research hotspot in academia and industry.Chatbot is constantly developing and expanding,the scope of its applications continues to expand,and people also have higher requirements for chatbot,hoping that while chatbot focuses on the content or task,it can also have more indepth emotional communication with people.This requires that chatbot can understand people's emotions and make reasonable emotional expressions.Therefore,emotional understanding,as the basis of reasonable emotional expression,has a very important research significance in chatbot.This paper focuses on the research of emotion recognition in chatbot.Unlike sentence-level sentiment classification in text sentiment analysis,the utterances in dialogue are not isolated.Therefore,emotion recognition in dialogue needs to consider more factors in dialogue and explore the role of these factors in emotion recognition.This paper mainly considers three different factors in dialogue,and designs three targeted dialogue emotion recognition methods to explore the actual impact of introducing these factors on the emotion recognition performance.The main research content includes:(1)For the contextual content information in dialogue,this paper mainly proposes a dialogue emotion recognition method based on hierarchical network,which models dialogue end-to-end through the utterance encoder and the context encoder to achieve the effective use of contextual content information and improve the performance of emotion recognition.(2)For the contextual emotion information in dialogue,this paper mainly proposes a dialogue emotion recognition method based on iterative improvement mechanism,which changes single-round prediction into iterative improvement multi-round prediction.This not only realizes the effective use of the existing contextual emotion information in the training phase,but also solves the problem that the contextual emotion information cannot be used in the test phase,and finally improves the performance of emotion recognition.(3)For the speaker information in dialogue,this paper mainly proposes a dialogue emotion recognition method based on mixed attention mechanism,which models various speaker emotional correlations by masked self-attention mechanism to achieve theeffective use of speaker information and effective improvement of emotion recognition performance.This paper attempts to consider three different dialogue factors in dialogue emotion recognition,designs three targeted methods to utilize them,and evaluates the proposed methods on two public datasets.Experimental results show that the three dialogue emotion recognition methods proposed in this paper have improved the performance of emotion recognition to varying degrees,proving the effectiveness of the proposed methods and the positive role of considering three dialogue factors.