Learning Emotion Recognition and Response Generation for a Service Robot

Abstract

Building dialoguing services for robots to provide natural human-robot interactions and to enhance user experiences is now advocated. With this type of services, a robot can work as a consultant and provide domain-specific knowledge to end users. In this study, we adopt a service-oriented framework to develop emotion-aware dialogues for a service robot. Our work includes several unique features: it trains classifiers to recognize users’ emotions in conversation, learns a deep neural model to generate answers in response to users’ questions, and uses the emotional information to determine the answer sentences produced by the dialoguing model. A series of experiments are conducted for performance evaluation. The results are compared with other machine learning methods, and they show the promise and potential of the presented approach.