| Abstract: A decision theoretic planner on instructional planning for intelligent tutoring systems (ITSs) is proposed. Three main difficulties on instructional planning problem, that are, an inaccuracy of student model, unreliability on effects of actions taken, and unexpected response from student, are modeled as probabilistic state transitions in Markov decision processes. The tutoring rules are considered as common pedagogy in the subject domain, and used to determine policies all over the states. Teachers' individual preferences of instruction, on the other hand, are modeled as the state reward function. A prototype ITS is implemented to evaluate the decision theoretic planner. It then is compared with a hierarchical task network planner implemented for the same task on the same subject domain. |
The project report is published in the workshop on Modeling Human Teaching Tactics and Strategies, ITS2000.
Noboru Matsuda, Kurt VanLehn, "Decision Theoretic Instructional Planner for Intelligent Tutoring Systems," in Proc. for Workshop on Modeling Human Teaching Tactics and Strategies, ITS2000, pp.72-83, 2000.
Download PDF version of the workshop paper
© Noboru Matsuda
Last modified: 11/27/04 15:35:27 -0500