Computational emotion recognition using multimodal physiological signals: Elicited using Japanese kanji words

This paper investigates computational emotion recognition using multimodal physiological signals. Four physiological signs - plethysmogram, skin conductance change, respiration rate and skin temperature - are measured to evaluate three emotions: positive, negative and neutral. Psychophysical experiments are conducted using Japanese kanji words in order to excite emotions in subjects so as to elicit physiological signals. For computational emotion recognition, machine-learning approaches, such as multilayer neural networks, support vector machines, decision trees and random forests, are used to design emotion recognition systems and their characteristics are investigated. In computational experiments conducted for recognising emotions, support vector machines equipped with a Gaussian kernel function attain a maximum averaged recognition rate of around 40% for all three emotions and around 56% for two emotions (positive and negative). The results obtained in this study shows that using multimodal physiological signals with a machine-learning approach is feasible and suited for computational emotion recognition.