Using Enhanced Concept Map for Student Modeling in a Model-Based Programming Tutor

We have been using the concept map of the domain, enhanced with pedagogical concepts called learning objectives, as the overlay student model in our intelligent tutors for programming. The resulting student model is fine-grained, and has several advantages: (1) it facilitates better adaptation during problem generation; (2) it makes it possible for the tutor to automatically vary the level of locality during problem generation to meet the needs of the learner; (3) it clarifies to the learner the relationship among domain concepts when opened to scrutiny; (4) the tutor can estimate the level of understanding of a student in any higher-level concept, not just the concepts for which it presents problems; and (5) two tutors in a domain can affect each other’s adaptation of problems. Prior evaluations have shown that tutors that use enhanced concept maps help improve learning. Student Modeling Traditionally, student models in tutoring systems have consisted of cognitive, affective and inferential components. The cognitive student model has been popularly built as an overlay of the domain model. Researchers have used various organizations for the domain model and the resulting overlay cognitive student model. These representations include conceptual graphs [5], Bayesian networks [22], directed acyclic graphs [10], tables [3] and Prolog clauses [19]. Conceptual graphs have been used because they are graphically inspectable, and facilitate interaction planning and student diagnosis [6]. Bayesian networks have been used to model cause and effect relationships among concepts (e.g., if you know 'for ' and 'while ' statements, you know loops). Tables and Prolog clauses provide a mechanism to aggregate concepts but do not explicitly represent any inherent relationships among them. Other organizations used for domain model include networks and trees. In WADEIn [2] used to teach student