Emotion Elicitation using Sequential Socio-Economic Decision-making in a Serious Game and Modelling Emotions with a Bayesian Approach

There is a growing body of research in the literature that investigates the relationship between emotions and decision-making in socio-economic contexts. Previous research has used in-person serious games based on Game Theory paradigms with socio-economic contexts to explore this relationship in controlled settings for both the neurotypical and autistic population. These games were predominantly applied in labs, which limited their applicability in scenarios such as, in-the-wild, for people on the autism spectrum and for people with affective disorders. This thesis investigates how a computerised serious game can be used to elicit specific emotions in its players. A digital serious game was developed that implemented socio-economic games from game theory as Non-Player Character (NPC) interactions and was deployed for the Android and WebGL platforms. The requirements for the game design were specified partly from the existing literature and partly from expert opinion. A co-design session was run with psychologists and therapists specialising in providing care to individuals on the autism spectrum to refine the game design. Three experiments were conducted, one in the lab and two in the wild. Hypothesis testing on the data showed statistically significant changes in self-reported emotion valence as per the intended design of the NPC interactions. These results show that, the NPC interactions could replicate the emotional responses in both settings that were previously just seen in labs with in-person gameplay and the players reported significant engagement during gameplay. This thesis then presents a probabilistic model of emotions from decision-making patterns seen from the players as a Dynamic Bayesian Network that outperforms a baseline model in predictive appropriateness. The findings from this thesis open a potentially new dimension of emotion detection using decision-making patterns which can be ubiquitously applied in-the-wild and in lab settings for several application areas, such as detecting early onset of affective disorders.