Emotion classification in human-computer-interaction on the basis of physiological data
This work addresses the still unsolved problem of stimulus- and subject-independent emotion identification from physiological data and presents, as a solution, a novel method for identification of affective changes from physiological data in the two dimensions of pleasure and arousal of the PAD (Pleasure-Arousal-Dominance) model. In our experiment, 110 participants passively viewed blocked affective pictures and then actively engaged in a mental training Wizard-of-Oz scenario while their physiological activity was recorded. In order to allow stimulus- and subject-independent testing, the group of subjects was split in two halfs. The first group’s IAPS picture viewing data was segmented into high pleasure, low pleasure, high arousal and low arousal physiological data and six ‘meaningful’ features were extracted from four physiological channels (EMG corrugator supercilii, EMG zygomaticus major, skin conductance, and blood volume pulse). These were used for creation of Emotion Identification Modules which were then applied on the second group’s affective picture viewing data for subject-independent emotion identification as well as on the first group’s Wizard-of-Oz scenario data for stimulus-independent emotion identification. With correct recognition rates of around 75 % (chance-level: 50 %) this method is a step towards a more accurate and robust stimulus- and subject-independent emotion identification.
Subject HeadingsAffective Computing [GND]
Human-computer interaction [LCSH]