Emotion Detection And Expression Dialog System

With the rapid development of natural language processing technology,human-computer interaction services integrating artificial intelligence have gradually become diversified on the Internet.The interaction between human and computer is not only the simple transmission of computer instructions,but also can simulate the way of information exchange between people to realize the interaction based on text and voice.As an important application of natural language processing technology in human-computer interaction,open domain dialogue system aims to carry out dialogue and interaction with people with unlimited subject content.Due to its wide application scenarios,it has become a research hotspot in recent years.The language signals of human communication include language rules such as word order and semantics,as well as implicit expressions such as emotion.At this stage,the open domain dialogue system does not integrate emotional perception into dialogue generation,but only focuses on understanding the semantic features of the text and generating grammatically correct responses.It lacks the ability to perceive users' emotions,and it is difficult to resonate with users emotionally in the dialogue.The current work of dialogue emotion perception is mainly based on the dialogue above.The understanding of the above can effectively solve the problem of polysemy in dialogue emotion perception.However,the existing research only takes into account the semantic features in the continuous dialogue,and does not take into account the real-time emotional state expressed by the user in his last speech,resulting in the loss of the user's short-distance emotional features.Therefore,this thesis proposes a dialogue emotion perception model integrating the semantic features of the dialogue and the emotional features of the user's recent speech,further integrates the emotion perception model into the dialogue generation task,and proposes a dialogue system integrating semantic understanding,dialogue emotion perception and emotion response.The main work and innovations of this thesis are as follows:1.In multiple rounds of dialogue,the dialogue system needs to perceive the emotional state of users in real time.Based on capsule network,this thesis proposes a dialogue emotion perception model integrating semantic features and emotional features.Compared with the traditional network using constants as neurons,the capsule model uses convolution operation to convert the dialogue text into multiple capsule vectors with depth,which can preserve more rich text features.The capsule model aggregates multiple capsule vectors into the feature expression of dialogue text at each emotional level through weight network and dynamic routing algorithm.Combined with the related work of dialogue emotion analysis in recent years,the experimental results of the emotion perception capsule model in multiple dialogue emotion analysis data sets have significantly improved the performance compared with the comparison model.2.The traditional dialogue generation model is based on the dialogue above.The semantic information of the dialogue above is embedded in the generation model to guide the model to generate the corresponding machine reply.The generation model based on dialogue above is vulnerable to the influence of general reply,and generates reply with strong generality but no practical significance.Therefore,this thesis introduces an emotion discriminator into the generation model to perceive the emotional information expressed by users in the conversation,and uses the emotional information to guide the generation model to generate responses with rich content and stronger emotional expression ability.At the same time,the attention mechanism is introduced between the encoder and the decoder,so that the decoder can pay attention to the useful information in the conversation when generating the reply.Finally,the effectiveness of the model is proved by comparative experiments.