"Yours Virtually", Advancing Affective Intelligence in Virtual Agents

As artificial intelligence increasingly enters social spaces, from classrooms to therapy apps, a central question emerges: can virtual agents understand not just what we say, but how we feel? This thesis introduces AvataRena, an Affective Intelligent Virtual Agent designed to interact in emotionally adaptive, socially aware, and human-like ways.

AvataRena bridges psychology and AI by embedding well-established psychological theories directly into its decision-making processes. By combining the Theory of Planned Behaviour, the Homophily principle, and Affect Control Theory, the agent is able to sense users’ emotional states, align with their social traits, and respond with context-appropriate verbal and non-verbal behaviour. The goal is not simply smarter AI, but more relatable, trustworthy, and emotionally intelligent interaction. Potential applications span mental health support, education, and marketing.

Three empirical studies evaluate AvataRena’s affective capabilities. The first study, involving 384 participants, demonstrates that users prefer virtual agents whose level of extroversion matches their own, confirming the power of homophily in human–AI interaction. The second study shows that integrating Affect Control Theory enables AvataRena to maintain emotionally coherent interactions, significantly improving social alignment compared to a non-affective control agent. The third study extends these findings into real-time conversations with 161 participants, where AvataRena consistently receives higher ratings for emotional experience and social perception.

Together, these results present a multidisciplinary framework for affective virtual agents that move beyond purely rational interaction. By integrating emotional theory, adaptive behaviour, and ethical safeguards, this thesis demonstrates how Intelligent Virtual Agents can evolve into socially attuned, emotionally responsive partners in everyday digital life.