Modeling Acceptable Novelty Based on Bayesian Information

Abstract

An individual’s acceptance of novelty depends on their emotions. We previously developed mathematical models of emotional dimensions associated with novelty, such as arousal (i.e., surprise) and valence (i.e., positivity and negativity). The models based on Bayesian theorem have three parameters: prediction error, uncertainty and external noise. Based on Berlyne’s arousal potential, we formulated valence as an inverse U shape function of arousal. We assume that the arousal level where the valence turns from positive to negative shows the range of novelty users would accept. In this study, we derive a corresponding prediction error and term this ‘acceptable novelty’. Our model predicts that the greater the uncertainty, the larger the acceptable novelty. Our experimental results using musical stimuli with novice participants supports the model prediction under assumption that uncertainty is greater than noise. By contrast, experts’ result is explained by the model prediction when uncertainty is as low as noise.