Using inverse planning for personalized feedback

An increasing number of automated models can make inferences about learners’ understanding based on their problem solving choices in interactive educational technologies. One potential use of these models is to personalize feedback interventions. We investigate using the output of an inverse planning model to choose feedback activities for learners. The inverse planning model uses the patterns of how a learner solves algebraic equations to estimate her proficiency on several discrete skills. The personalized feedback then focuses on the skill which is least proficient and includes a combination of existing educational content and scaffolded practice. We experimentally tested the effectiveness of personalizing the feedback based on the algorithm’s estimate compared to simply providing a random feedback activity. The results show that completing the feedback was associated with performance improvements from pre- to post-test, but that personalized feedback was not associated with reliably more improvement. However, participants who received feedback about a skill that was far from mastery did show reliably more improvement than those who received feedback about an already-mastered skill. This suggests that there is potential in using the inverse planning algorithm to provide more effective learning experiences.