In this paper, we present work to construct a robotic tutoring system that can assess student knowledge in real time during an educational interaction. Like a good human teacher, the robot draws on multimodal data sources to infer whether students have mastered language skills. Specifically, the model extends the standard Bayesian Knowledge Tracing algorithm to incorporate an estimate of the student's affective state (whether he/she is confused, bored, engaged, smiling, etc.) in order to predict future educational performance. We propose research to answer two questions: First, does augmenting the model with affective information improve the computational quality of inference? Second, do humans display more prominent affective signals in an interaction with a robot, compared to a screen-based agent? By answering these questions, this work has the potential to provide both algorithmic and human-centered motivations for further development of robotic systems that tightly integrate affect understanding and complex models of inference with interactive, educational robots.
[1]
Vincent Aleven,et al.
More Accurate Student Modeling through Contextual Estimation of Slip and Guess Probabilities in Bayesian Knowledge Tracing
,
2008,
Intelligent Tutoring Systems.
[2]
Yanbo Xu,et al.
EEG Helps Knowledge Tracing !
,
2014
.
[3]
Ayanna M. Howard,et al.
Applying Behavioral Strategies for Student Engagement Using a Robotic Educational Agent
,
2013,
2013 IEEE International Conference on Systems, Man, and Cybernetics.
[4]
Bilge Mutlu,et al.
Pay attention!: designing adaptive agents that monitor and improve user engagement
,
2012,
CHI.