Evaluating emotional algorithms using psychological scales

Abstract

Evaluating the effectiveness of emotional algorithms is a challenge in the research of affective computing. This paper presents an evaluation method using psychological scales. Affective interfaces and artifacts generated by different emotional algorithms are regarded as the external stimuli, while users' feelings are regarded as subjective evaluations. Psychophysical methods, such as the Scheffe method of paired comparison, are adopted to measure users' feelings and obtain a psychological scale. Thus objects generated by different algorithms can be compared on this psychological scale. The higher the value, the better the algorithm is. Empirical studies on three kinds of emotional algorithms, namely emotional image retrieval, emotional speech synthesis, and emotional music generation, are used to illustrate various uses of our approach and how it may be applied to evaluating affective interaction and user-centered design.