Research On Emotion Guidance Mechanism For Multi-turn Dialogue Generation

Dialogue generation is an important branch in the field of natural language processing.It mainly studies generative dialogue models to generate responses close to the human level.Emotion plays an important role in dialogue generation,especially in multi-turn dialogue generation.The emotional appropriateness of generated responses seriously affects the sustainability of human-computer dialogue interaction and the user experience of the dialogue system.To improve the emotional appropriateness of generated responses and achieve the purpose of user's emotion guidance,it is a relatively direct research idea to design an emotion guidance mechanism in the multi-turn dialogue generation model,and dynamically control the generation of responses from the emotional level.This research helps to improve the emotional intelligence of the dialogue system and promotes the practical application of research on multi-turn dialogue generation in human life.The specific research content and methods are as follows:Firstly,to solve the problem of the accuracy of emotional decision-making,the multiturn dialogue emotion guidance decision-making model was researched,and the emotion guidance decision-making cell was proposed.The model simulates the emotional dialogue interaction scenarios between people and is improved based on LSTM,adding an emotion guidance gate to control the degree of influence of the user's guiding target sentiment on the current emotional decision-making.The experimental results of the multi-turn dialogue emotion guidance decision-making task show that the model can finally decide the emotion of response dynamically according to the specified guiding target sentiment polarity and historical emotion sequence,and realize the improvement of the accuracy of emotional decision-making.Secondly,to solve the problem of emotional consistency in generated responses,research was carried out on the response generation model of specified emotion,and the multi-turn dialogue emotional response generation unit and emotion reranking strategy were proposed.The model uses GPT-2 as the basic model for multi-turn dialogue generation,uses a multi-turn dialogue corpus with emotional tags to fine-tune the model,and combines an emotional reranking strategy to achieve emotional controllable dialogue response generation.The experimental results of the multi-turn dialogue emotion response generation task show that the model can finally generate the response of the specified emotion based on the dialogue context,which effectively improves the emotional consistency of the generated response.Finally,to solve the problem of emotional appropriateness of generated responses and achieve the purpose of user's emotion guidance,the multi-turn dialogue emotion guidance response generation framework was researched,and the emotion guidance response generation model was proposed.The model organically combines an emotion guidance mechanism including an emotion classifier,an emotion guidance decision-making cell,and an emotional response generation unit,and dynamically controls the generation of emotional response in the interactive process of multi-turn dialogue to realize the guidance of the user's emotion.The experimental results of the multi-turn dialogue emotion guidance response generation task show that the model can finally generate a response that guides the user's emotion according to the dialogue history and the specified sentiment polarity of the guiding target,and has a certain ability to guide the user's emotion.