Research On Dialogue Generation Methods Fusing Emotional Cognition

With the development of artificial intelligence technology,people's expectations for dialogue systems have shifted more to the need for communication.Emotion is an important factor affecting interpersonal communication.The ability to have emotional cognition and expression is a higher level of intelligence,which can understand and meet human needs from a deeper level.However,most of the current research on dialogue generation is dedicated to improving the diversity and fluency of replies,ignoring the requirements of emotional expression.Research on dialogue generation methods fusing emotional cognition takes the emotional information in the dialogue as an entry point,by predicting and characterizing the emotional information in the dialogue text,the dialogue system has the ability of emotional perception,and then introduces the emotional information into the dialogue generation process to guide the dialogue system to generate empathy responses.Mainly done the following two aspects of work:Construction of sentiment analysis model for dialogue text.Aiming at the problem that semantic information is difficult to mine due to the dialogue text is noisy,the length is short,and the complexity of information,a sentiment analysis architecture named Multi-LG based on the fusion of local semantic features and global semantic features is proposed.This architecture can effectively extract emotional semantic features in dialogue text.Specifically,the local semantic features in the dialogue text are fully mined through the convolutional neural network,and the global semantic features in the dialogue text are fully mined through the bidirectional long short-term memory network based on the self-attention mechanism,and take the fusion feature of the two as the emotional semantic feature of the dialogue text.By comparing experiments with multiple models on the Weibo corpus,the effectiveness of the architecture is proved.The construction of dialogue generation model fusing emotional cognition.Aiming at the problem of lack of emotion in the existing dialogue system,an Emotional Dialogue Generation Model(Emo DG)is proposed.According to the Multi-LG model,the emotional semantic features are obtained as the emotional vectors of the dialogue text.The emotional vectors are embedded in the encoding stage,the emotional vectors are injected at each time step in the decoding stage,the emotional distance penalty isintroduced into the loss function,and the candidates are based on the emotional distance.The sequence is reordered so that the model can generate replies that are not only semantically fluent but also have specific emotional expressions.A comparative experiment on Qingyun corpus proves that the model enhances the ability of emotional expression on the basis of fluent semantic expression and can generate empathic responses.