Using Knowledge Components for Collaborative Filtering in Adaptive Tutoring Systems

In adaptive tutoring systems, accurately assessing the ability of a student is central to prescribing the tasks that best facilitate learning. For the 2010 KDD Cup challenge a data set of logs from the Cognitive Tutor system was made available, and contestants were asked to predict the correctness of a student’s attempt to answer questions. A successful approach included a collaborative filtering system which predicted student performance on the basis of the performance of similar students. In this paper, we present an extension of this approach. Rather than finding similar students on the basis of their performance on specific questions, we based our similarity measure on the performance on questions that require the same “knowledge components” (or skills). This approach increases the amount of users with whom it is possible to compare performance, which in turn increases the likelihood of finding similar students. The experiments using our question type-based distance measure yield promising results.