Subgroup Discovery with User Interaction Data: An Empirically Guided Approach to Improving Intelligent Tutoring Systems

Learner modeling, a challenging and complex endeavor, is an important and oft-studied research theme in computer-supported education. From this perspective, Educational Data Mining (EDM) research has focused on modeling and comprehending various dimensions of learning in computer based learning environments (CBLE). Researchers and designers are actively attempting to improve learning systems by incorporating adaptive mechanisms that respond to the varying needs of learners. Recent advances in data mining techniques provide new possibilities and exciting opportunities for developing adaptive systems to better support learners. This study is situated in the context of clinical reasoning in an Intelligent Tutoring System called BioWorld and it aims to examine the relationship between the lab-tests ordered and misconceptions held by learners. Toward this end, we employ an EDM technique called subgroup discovery to unpack the rules that embody the hypothesized link. Examining such links may have implications for identifying the points along learning trajectories where learners should be provided the requisite scaffolding. This study represents our efforts to evaluate and derive empirically based design prescriptions for improving Intelligent Tutoring Systems. Implications for practice and future research directions are also discussed.