Adding Weights to Constraints in Intelligent Tutoring Systems: Does It Improve the Error Diagnosis?

Abstract

The constraint-based modeling (CBM) approach for developing intelligent tutoring systems has shown useful in several domains. However, when applying this approach to an exploratory environment where students are allowed to explore a large solution space for problems to be solved, this approach encounters its limitation: It is not well suited to determine the solution variant the student intended. As a consequence, system's corrective feedback might be not in accordance with the student's intention. To address this problem, this paper proposes to adopt a soft computing approach for solving constraint satisfaction problems. The goal of this paper is two-fold. First, we will show that classical CBM is not well-suited for building a tutoring system for tasks which have a large solution space. Second, we introduce a weighted constraint-based model for intelligent tutoring systems. An evaluation study shows that a coaching system for logic programming based on the weighted constraint-based model is able to determine the student's intention correctly in 90.3% of 221 student solutions, while a corresponding tutoring system using classical CBM can only hypothesize the student's intention correctly in 35.5% of the same corpus.