Dynamic Cognitive Tracing: Towards Unified Discovery of Student and Cognitive Models
José P. González-Brenes, Jack Mostow · 2012
This work describes a unified approach to two problems pre-viously addressed separately in Intelligent Tutoring Systems: (i) Cognitive Modeling, which factorizes problem solving steps into the latent set of skills required to perform them [7]; and (ii) Student Modeling, which infers students ’ learn-ing by observing student performance [9]. The practical importance of improving understanding of how students learn is to build better intelligent tutors [8]. The expected advantages of our integrated approach include (i) more accurate prediction of a student’s future perfor-mance, and (ii) clustering items into skills automatically, without expensive manual expert knowledge annotation. We introduce a unified model, Dynamic Cognitive Trac-ing, to explain student learning in terms of skill mastery over time, by learning the Cognitive Model and the Stu-dent Model jointly. We formulate our approach as a graph-ical model, and we validate it using sixty different synthetic datasets. Dynamic Cognitive Tracing significantly outper-forms single-skill Knowledge Tracing on predicting future student performance. 1.