Intelligent tutoring systems (ITSs) are used increasingly to support online learning processes in different disciplines. To be able to provide a student with effective feedback within an ITS, the granularity of the student’s input plays an important role: the more coarse-grained the input is, the more difficult it is to provide specific feedback. This practice-oriented research seeks to identify student strategies and errors based on minimal input. For mathematics, we propose to develop means to determine students’ errors based on coarse-grained input through model backtracking (MBT). With MBT, a task is modelled such that different errors can be distinguished in student answers.
The main research question is: How can online learning processes be improved though error identification in coarse-grained input using MBT in Grade 11 ‘havo wiskunde A’? We start with a design study implementing MBT for two cases, followed by testing in teaching experiments. We use two task-based interviews with ‘havo wiskunde A’ students (n=25 each) to study the validity of error detection trough MBT, improving the initial design. We then scale up towards testing in a statistical study in Grade 11 ‘havo wiskunde A’.
We use an experimental group receiving specific feedback through MBT and a control group (n=100 each). We use datamining on the interaction logs to catalogue different learning strategies together with an analysis of variance on pre- and posttest scores to determine whether MBT contributes to positive online learning experiences and outcomes for students.