Estimation of Emotional Evoked State Based on Image Processing of Advertisement Images and CTR Prediction

Abstract

The decision to click on an advertisement image depends on several factors. In this study, we analyzed the relationship between the emotional arousal indicators of images and Click-Through-Rate (CTR), focusing on the impression that advertisement images have. We introduced two emotional arousal indicators, valence which is evaluated as positive and negative, and arousal which is evaluated as low and high. To predict valence and arousal values of advertisement images, we developed an emotional arousal indicator estimation model using image sets containing normative ratings on the two emotional dimensions of valence and arousal. In addition to the valence and arousal predicted by this model, we included metadata associated with the advertisement images and thus saw the direct effect of valence and arousal on CTR.