Probability Estimation And Competence Models For Rule And Strategy-basedE-TutoringSystems
Room: BBL 165
Speaker: Diederik Roijers
Title: Probability Estimation and Competence Models for Rule and Strategy-based e-Tutoring Systems (thesis defense)
Many problems in natural sciences can be solved by rewriting a formula step by step, to a solution. Systems like ideas, so called rule and strategy based e-tutoring systems, can help students study by automatically providing feedback. For these systems we have created a student model. The student model makes the rule properties (difficulty and discriminativity) and student properties (start competence and learning speed) operational, by relating them to the probability of a student correctly applying a rewrite rule. In this research we provide parameter estimation procedures, show that these are reliable using simulations, and test our model experimentally. In the experiments we use real-life data from students from three different groups. The students do excersises in rewriting a given logic formula to disjunctive normal form. We asked domain experts, teachers at universities, to rank the difficulties of the rewrite rules in the logic domain. We found that these
rankings correpond fairly well to the parameter estimates we obtain experimentally. We also show that the values for start competence
are what we expected based on our knowledge of the different student groups. Discriminativity and learning speed could not be
estimated well because of too little data. The student model we create is a promising model, which calls for further research.