Computational Models of Players´ Physiological-Based Emotional Reactions: A Digital Games Case Study

Emotionally adaptive games are one of the holy grails of modern affective game research. However, current state of the art affective games rely on static game adaptation mechanics that assume a fixed emotional reaction from players every time. Not only this, most commercial titles have no affective adaptation loop whatsoever and their design is based on game design optimizations via typical beta-testing procedures, which falls short of ideal both in the level design and long-term game play experience fronts. In this paper, we demonstrate a generalizable approach for building predictive models of players´ emotional reactions across different games and game genres. We describe a physiological approach for modelling players´ emotional reactions, which relies on features extracted from players´ emotional responses to game events, which were collected and extrapolated through their physiological data during actual game play sessions. Based on the optimal feature sets found by three feature selection algorithms (best first, sequential feature selection and genetic search), the collected features are used to create computational models of players´ emotional reactions on the arousal and valence dimensions of emotion, using several machine learning algorithms. We expect this approach will allow both a more objective and quicker prototyping for digital games, as well as foster a future generation of affective games capable of modelling players´ affective profiles over time, thus adapting to their changing preferences and needs.