The Case for Cognitive-Affective Architectures as Affective User Models in Behavioral Health Technologies

As technology proliferates into all aspects our lives, affect-adaptive interaction becomes increasingly important. A growing area where this is particularly critical is behavioral health technology: mental health (MH) apps, serious therapeutic games, virtual agents and social robots, and virtual reality environments. These technologies aim to support mental health and wellness via psychoeducation, behavior coaching, supportive and motivational interventions, and opportunities to practice coping skills. Affect-adaptive interaction requires detailed affective user models, capturing the range of the user's affective states, their triggers, expressive manifestations, and consequences on behavior, within the specific interaction context. Cognitive-affective architectures have been proposed as a possible approach to affective user modeling, due to their ability to represent detailed information about internal processing and support complex what-if and abductive reasoning. However, their construction is highly labor-intensive, requiring significant knowledge engineering and parameter tuning. In this paper I argue that the unique interaction context offered by MH apps helps address these issues. By facilitating self-reporting and non-intrusive sensing, mobile apps support user-driven construction of individual- and context-specific cognitive-affective architectures. The focus on emotions makes the explicit collection of affective data central, thereby encouraging direct user involvement, which is also enhanced by engaging the user in the on-going architecture refinement. I describe an approach to constructing a cognitive-affective architecture-based affective user model, providing several illustrative examples which demonstrate how the model would support more personalized assessment, intervention and progress tracking in mental health apps. The paper concludes with a discussion of challenges and opportunities.