The work aims to improve the user comfort measurement method in human-robot interaction and enhance the intelligent perception ability of robots. Based on a robot massage operation platform, measurement data, including the force and deformation of the skin during human-robot interaction, as well as subjective perception evaluation were collected. A three-dimensional comfort perception space based on pain, stretching, and burning sensation was built. Then a mapping relationship between objective measurement data and human perception was established through the back propagation neural network. The support vector machine algorithm was used for comfort classification of human. Results showed that the perception prediction model and state classification model can effectively achieve overall identification from feature information to human perception, and then to comfort identification. This method can be used to connect and recognize the operation characteristics and the subjective perception of human body in human-robot interaction, and provide an effective basis for safe and comfortable operation planning and online control of robots.