Affect Prediction for Emotional Design: A Comparison Study of Physiological and Subjective Self-Report Data

Abstract

Emotional design has attracted much attention due to its important role in the development of products and services towards high value-added user satisfaction and performance enhancement. However, how to predict users’ affective states in real time and without having to interrupt the user is critical to emotional design. This study compared affect prediction between using physiological measures and using self-report subjective measures. Specifically, an experiment was designed to elicit seven different affective states using standardized affective pictures as visual stimuli. Each stimulus was presented for 6 seconds and multiple physiological signals were measured, including facial electromyography, respiration rate, electroencephalography, and skin conductance response. Subjective ratings were also recorded immediately after stimulus presentation. Three data mining methods (i.e., decision rules, k-NN, and decomposition tree) based on the rough set theory were applied to construct prediction models from physiological measures and subjective measures, respectively. We obtained the highest mean prediction rate at 73.69% for physiological models and 52.43% for subjective models, respectively, across the 7 affective states. It demonstrates that physiological data are able to predict better result than subjective self-report data did and that physiological computing offers great potential for the development of emotional design.