View article

A critical part of domain modeling in intelligent tutoring systems (ITSs) is to determine how the domain content relates to the skills and knowledge that we aim for the student to learn about. This task can be achieved through data-driven techniques. If domain content is associated with a set of questions or exercises, student performance data over these items can be used to find the latent skills behind the content.

Data-driven techniques that map items to skills fall within two main categories:(1) entirely driven from student data performance or (2) starting with expert-given mapping and refined based on these data. Entirely data-driven techniques are appealing because they dispense the tedious efforts required to do the mapping by content experts. However, the mapping obtained from such methods may be hard to interpret and will almost surely contain latent skill factors that do not match the pedagogical structure of the learning content. For this reason, the refinement of expert given mappings has greater utility in most contexts and this chapter focuses on this specific problem.