Novel online tutor modeling for intelligent tutoring systems

Abstract

Example tracing tutors has proven to be one of the simplest methods for modelling domain limited scenarios within intelligent tutoring systems [1]. The intelligence embedded within this type of tutors lies in the feedback it possesses during creation and how it is triggered based on learner interaction with the example. Since real-life tutoring in 1-to-1 scenarios proves to have the ultimate learning gain factor of 2 sigma [7]; the tutor model planted in example tracing is very limited compared to real life tutor behavior. In this paper we present a novel tutor modeling technique that records instructional behavior and scaffolding scenarios on top of example tracing and student responses. The model is designed conforming to the standard teacher and student dialogue moves [8] and in a way that allows evolving the information in the model while it is being used. A set of 3 face to face lectures each with 10 problems in math, has been logged and analyzed to extract the principle moves of the instructor strategies and related keywords. A special virtual classroom was developed for simultaneous capturing of teacher dialogue moves and additional VCR tools, along with student responses to construct the model information.