Towards Assessing Students' Prior Knowledge from Tutorial Dialogues.
Dan C. Stefanescu, Vasile Rus, Arthur C. Graesser · 2014
This paper describes a study which is part of a project whose goal is to detect students ’ prior knowledge levels with respect to a target domain based solely on characteristics of the natural language interaction between students and a state-of-the-art conversational Intelligent Tutoring System (ITS). We report results on dialogues collected from two versions of the intelligent tutoring system DeepTutor: a micro-adaptive-only version and a fully-adaptive (micro- and macro-adaptive) version. We extracted a variety of dialogue and session interaction features including time on task, student-generated content features (e.g., vocabulary size or domain specific concept use), and pedagogy-related features (e.g., level of scaffolding measured as number of hints). We present which of these features are best predictors of pre-test scores as measured by multiple-choice questions.