Principal Investigator: James Lester (PI, Computer Science)
Primary Participants (Computer Science): Joe Grafsgaard (Computer Science), Scott McQuiggan (Computer Science), Jen Robison (Computer Science)
Sponsor: National Science Foundation – Human-Centered Computing Program (2008-2011)
Objectives: Because of the growing recognition of the role that affect plays in learning, affective computing has become the subject of increasing attention in research on interactive learning environments. Game-based learning environments present a significant opportunity to investigate student affect in interactive learning. In particular, narrative-centered game-based learning environments, which are game-based learning environments where learning activities play out in dynamically generated interactive narratives and training scenarios, afford great opportunity for exploring computational models of student affect. Leveraging the rich interactions and motivational strengths offered by narrative-centered learning environments, the project has two major thrusts:
1. Develop a full suite of student affect modeling technologies for game-based learning environments. To create highly motivating learning interactions and to thus promote effective learning, we are creating an inductive computational framework for student affect modeling in narrative-centered game-based learning environments. During the training phase, the affect modeler monitors student affective responses as they solve problems in a narrative-centered learning environment. As the interactive narrative plot unfolds, the affect modeler also monitors student physiological signals (e.g., galvanic skin response, heart rate, and posture), their interactions with the cast of virtual characters, and their problem-solving actions. After inducing affect models from the logged interactions using an array of machine learning methods (decision trees, naïve Bayes classifiers, n-grams, support vector machines, Bayesian networks, Hidden Markov models), the deployed affect modeler will at run-time use the models to 1) perform affect recognition, in which it will predict students’ affect states, recognize engagement and flow, detect frustration, and monitor engagement, and 2) perform affect expression, in which it will customize pedagogical activities and dynamically plan the motivating, empathetic responses of the virtual agents in the learning environment. All design, implementation, and evaluation activities are being carried out in Crystal Island, a narrative-centered game-based learning environment for biology.
2. Provide a comprehensive account of both the cognitive-affective processes of students interacting with affect-informed game-based learning environments and the results of such interactions. To understand the cognitive-affective mechanisms by which learning occurs in affect-informed game-based learning environments, we are taking an empirical approach to investigating student affect modeling. For affect recognition, the studies are investigating motivation, engagement, flow, and frustration. For affect expression, they are investigating the virtual agent affect support strategies of mastery learning, vicarious learning, motivation, and integrated efficacy-affect support. Empirical studies of the computational models will assess the approaches to affect recognition and affect expression that best address the requirements for accuracy, early prediction, efficiency, and robustness. Human subjects studies will determine precisely which affect modeling techniques best close the “affective loop” and contribute most effectively to student learning effectiveness and motivation.
