A New Method for Measuring Similarity Between Educational Items from Response Data

A main goal of Educational Data Mining (EDM) is developing methods for exploring large-scale data that come from interactive learning environments, and using those methods for improving learning outcomes [1] • A fundamental question within EDM (and Psychometrics) is identifying groups of items (questions) that require the same set of skills. • Standard statistical methods that are used for that are based on the assumption that student’s performance on items that require the same skill should be similar (see for example in [2]). This holds if the latent trait is relatively fixed during the activity being measured, as in the context of testing • However, this assumption does not hold in the context of learning, which means that the latent trait changes rapidly • We propose a novel similarity measure, termed Kappa Learning, which aims to identify similarity between items under the assumption that the latent trait can change, namely, that students can acquire new skills during the activity