AbstractIn computerized tutoring, the pace of instruction is related to the student's mastery levels of the learning objectives. The observable student's behavior that can be used to measure his knowledge is usually his responses to test items. Unobservable variables that are related to learner's motivation can affect learning but are difficult to quantify. In comparison with other decision-theoretic tutoring systems, the novelties of this research are: (1) the efficiency-centric approach to develop the Bayesian networks; (2) the formulation of utility values for different tutoring outcomes that are independent of past actions and to satisfy the separability condition; (3) the development of a common measure for student's mastery levels and item difficulties; and (4) the generation of optimal policies in polynomial time. A prototype web-based tutoring system, known as iTutor, incorporating the novelties has been developed for engineering mechanics. Formative evaluations of iTutor have shown encouraging results.
Student model, Bayesian network, decision analysis, item selection