Intelligent Tutoring Systems

Abstract

Intelligent Tutoring Systems (ITS) are meant to provide useful tutoring services for assisting the student. These services include coaching, assisting, guiding, helping, and tracking the student during problem-solving situations. To offer high-quality tutoring services, an ITS must be able to establish the correct student profile, then understand and diagnose the student cognitive as well as its affective state. This special issue of Educational Technology & Society presents recent works dealing with those matters. Extracting Procedural Models Using Educational Data Mining The main goal of an intelligent tutoring system is to actively provide guidance to the student in problem-solving situations. Relevant feedback should be founded on a thorough understanding and diagnosis of student responses. Building such understanding and diagnosis model is a difficult issue that is also a time-intensive process involving human experts. This issue becomes even more difficult in ill-defined domains where an explicit representation of the training task is hard, if not impossible, to set up. Educational data-mining (EDM) brings some promising solutions to this issue. You will find in this special issue two EDM-based solutions proposed for coping with this problem. Each of these solutions consists of a model that can constantly learn from new learner or user data and thus, guaranties that the tutor provides an up-to-date feedback. In one hand, Barnes and Stamper propose a novel application of Markov decision processes (MDPs) to automatically generate hints for an intelligent tutor that learns. This approach eases the process of building the understanding and diagnosis model of student actions. The authors extracted MDPs from four semesters of student solutions created in a logic proof tutor, and calculated the probability of being able to generate hints for students at any point in a given problem. The results indicate that extracted MDPs and their proposed hint-generating functions are able to provide hints over 80% of the time. The results also indicate that they can provide valuable tradeoffs between hint specificity and the amount of data used to create an MDP. In the other hand, Fournier-Viger et al. present a novel framework for adapting the behavior of intelligent agents based on human experts' data. The framework consists of an extended sequential pattern-mining algorithm that, in combination with association rule discovery techniques, is used to extract temporal patterns and relationships from the behavior of human learners of multiple profiles, executing a procedural task. The proposed framework has been integrated within CanadarmTutor, an intelligent tutoring system aimed at helping students solve procedural problems that involve moving a robotic arm in a complex virtual environment. CanadarmTutor acts in an ill-defined domain where the problem space associated with a given task consists of an infinite number of paths. The framework was used to improve the behavior of a cognitive agent that adapts its decision by learning from data gathered during past cognitive cycles. …