Modeling Emotions as Latent Representations of Appraisals

Emotion recognition is usually achieved by collecting features (physiological signals, events, facial expressions, etc.) to predict an emotional ground truth. This ground truth, however, is subjective and not always an accurate representation of the emotional state of the subject. In this paper, we show that emotion can be learned in the latent space of machine learning methods without relying on an emotional ground truth. Our data consists of physiological measurements during video gameplay, game events, and subjective rankings of game events for the validation of our hypothesis. By calculating the Kendall ${\tau}$ rank correlation between the subjective game event rankings and both the rankings derived from Canonical Correlation Analysis (CCA) and a simple neural network, we show that the latent space of these models is correlated with the subjective rankings even though they were not part of the training data.