Design And Implementation Of Dialogue System Based On Emotion-Enhanced

As the research on human-computer dialogue becomes more and more mature,the application of human-computer dialogue system has gradually penetrated into people’s life and work.With the development of research and development,people’s demand for open-domain dialogue systems to be rich in emotion is also increasing.Although a lot of research is currently considering the emotional embedding of dialogue systems,there are still existing emotional open-domain human-machine dialogues.There are many problems such as confusion in emotional perception,incoherent dialogue language,and imperfect emotional expression.Therefore,it is of great significance to study and solve these problems for emotional open-domain dialogue systems.Relying on the special topic of "Key Technologies of Human-like Intelligent Perception and Emotional Interaction Driven by Big Data",this thesis proposes to assist the system’s perception of user emotions and the generation of emotional dialogues by extracting information from multiple data such as dialogue behaviors and dialogue topics in multiple rounds of dialogue.,a human-machine dialogue system for emotional enhancement.The main research contents of this thesis are as follows:(1)The relative position-based multivariate data encoding is designed and implemented.When encoding dialogue texts,emotions,and behaviors,due to the limitation of the absolute position encoding method on the length of sequence processing,this thesis proposes a method of multivariate data encoding using relative dialogue information to capture the relationship between words in long sequences.In addition,a finite coded value is used to represent a position of infinite length,which improves the multi-round dialogue capability of the dialogue system.(2)Design and implement a Transformer-based multi-task emotional dialogue generation model.In general,emotion perception and dialogue generation are separate tasks.Emotion recognition can only pass the recognized emotion results to the dialogue generation part,but the information concerned in the recognition process cannot be transmitted.Moreover,the dialogue process not only contains emotions,but also includes Therefore,the multi-task learning method is adopted in this thesis to share the parameters of emotion prediction task,behavior prediction task and dialogue generation task through hard constraints to assist emotion perception and emotion expression.(3)Design and implement an emotion-enhancing human-machine dialogue system.It mainly includes a client module,a server module and a dialogue generation module.The dialogue generation module is the above-mentioned emotional dialogue generation module based on multi-task learning using relative position coding,and realizes a visual interaction interface so that users can interact more smoothly.