Adaptive task selection in automated educational software: A comparative study

In this study, we consider the challenge of adapting the difficulty level of the tasks suggested to a student using an educational software system. We investigate the effectiveness of different learning algorithms for the challenge of adapting the difficulty of the tasks to a student’s level, and compare their efficiency by means of simulation with virtual students. Our results demonstrate that the methods based on Bayesian inference outperformed most of the other methods, while in dynamic improvement domains, the item response theory method reached the best results. Given the fact that correctly adapting the tasks to the individual learners’ abilities can help them increase their improvement and satisfaction, our study can assist the designers of intelligent tutoring systems in selecting an appropriate adaptation method, given the needs and goals of the educational system, and given the characteristics of the learners.