Learning Optimal and Personalized Knowledge Component Sequencing Policies

One of the goals of adaptive learning systems is to realize adaptive learning sequencing by optimizing the order of learning materials to be presented to different learners. This paper proposes a novel approach to recommending optimal and personalized learning sequences for learners taking an online course based on the contextual bandit framework where the background knowledge of the learners is the context. To improve learning efficiency and performance of learners, the adaption engine of such an adaptive learning system can select an optimal learning path for a learner by continually evaluating the learners’ progress as the course advances. To overcome the complexity of learning path recommendation due to the large number of knowledge components, we use the ‘divide-and-conquer’ approach to modeling the domain and designing the sequence adaptation algorithm. Also, the adaptation engine can dynamically replan the learning path for a learner if her/his performance is worse than expected. Finally, our approach can improve over time by learning from the experience of previous learners who adopted recommended sequences.