Learner Modeling to Predict Real-Time Affect in Serious Games

In recent decades research on human emotion has revealed that affect plays a central role in human cognition. Historically, emotion was viewed as separate from cognition, and only recently has there been consensus that affect is a central component of rational behavior and social interaction (Forgas, Wyland, & Lahan, 2006). These findings have motivated the goal of designing affect-sensitive computer systems capable of recognizing and responding to users’ emotional states. This has led to the development of affective systems with a wide variety of capabilities and purposes, such as therapeutic virtual agents, mood detection of large social network groups, and empathetic tutorial systems (Picard, 1997). Applications such as these require that three critical decisions must be made to inform how affect will be modeled and detected: 1) How will affect be represented? 2) What signals will be used for affect detection? 3) How will affect predictions be utilized?