Research And Design Of Service Robot Dialogue System Based On Emotional Interaction

In recent years,problems such as aging of population have become increasingly prominent.The urgent needs of society and the rapid development of artificial intelligence and robot technology have greatly promoted the research,development and application of all kinds of service robots.Voice and dialogue technology can provide the most humanized way of interaction for service robots.With the rapid development of artificial intelligence technology such as deep learning,the generative dialogue algorithm of intelligent robots has better generalization and practical application value.In this paper,based on the application requirements of a family escort robot independently researched and developed by the laboratory,a set of emotion-based service robot generation dialogue system is designed by deep neural network,which is encapsulated into cloud service for service robot call.The difficulty of dialogue system is to generate natural,reasonable,interesting and diverse responses based on questions after speech recognition.Therefore,it is necessary to first understand people's needs,that is,to distribute user input to the correct scene,then generate a reasonable response.This paper summarizes the research status of scene distribution and dialogue generation related technologies,and analyzes the requirements of the service robot dialogue system and the overall design of the system.Detailed research and detailed design of scene distribution and dialogue generation are carried out.Scene distribution is essentially a text classification problem.Because different domains have different processing logic,accurate scene distribution is very crucial.In this paper,the classical convolution neural network is selected as the basic classification network,and the input word vector is transformed into Chinese character-level vector input.The input character-level vector is updated dynamically during the training process,and the classification effect is obviously better than the static word vector.In view of the inaccuracy or incompatibility of the single model classification in the actual chat conversation application,this paper combines the two methods of the character level convolution network classification and the word vector similarity classification,and proposes a new classification model.The accuracy of the fusion classification algorithm is proved by the experiment.The dialogue module realizes the open domain chat question and answer,that is,the "chat" category in the scene distribution,which is used to process chat scenes that do not belong to the specified categories.The goal is not to complete the task assigned by the user,but to attract the users to communicate.This paper proposes an improved emotion-based generative end-to-end dialogue model for emotional dialogue model ECM which needs to artificially set emotions,and an emotional classifier is added to generate the response of the corresponding emotion.In the prediction generation stage,a variety of cluster search method with emotional factors is used to make the generated response more vivid and reasonable.A large number of artificial evaluation results show that the model of emotional dialogue generation designed in this paper is more superior.