Investigating the utility of a conjunctive model in Q-matrix assessment using monthly student records in an online tutoring system
Nathaniel O. Anozie, Brian W. Junker · 2007
AbstractHeffernan et al. (2001) have developed an online intelligent tutoring system(the ASSISTmentsystem) for eighth grade mathematics that is explicitly aligned tostate exam standards. In this paper we assess the utility of the DINA model (Junkerand Sijtsma, 2001) for diagnosing knowledge components (KC’s: skills, piecesof knowledge, and other cognitive attributes) that students do and do not have,based on their interactions with the tutoring system. In a split-half cross-validationexperiment we found that with 50 or more questions per student we could obtain70% prediction accuracy. We also found that many students had attained 30%or less of the KC’s associated with these questions in a Q-matrix based transfermodel. As expected, the KC’s students lack are not basic whole-number, addition,and similar skills, but higher-order symbolic skills such as algebraic manipulationand equation solving. Overall, we find that the DINA model is q uite promising foron-line KC-based diagnosis and reporting.keywords: diagnostic assessment, benchmark assessment, Bayes Net, intelligenttutor