Episodic Learner Modeling

Modeling the learner is a central aspect of intelligent tutoring systems and knowledge-based help systems that support learners in complex problem-solving domains. In this article, the episodic learner model ELM is introduced as a hybrid system that analyses novices' solutions to programming tasks based on both rule-based and case-based reasoning. ELM behaves like to a human tutor. Initially, ELM is able to analyze problem solutions based only on its domain knowledge. With increasing knowledge about a particular learner captured in a dynamic episodic case base, it adapts to the learner's individual problem-solving behavior. Two simulation studies were performed to validate the system. The first study shows that the system can learn which rules are applied successfully to diagnose code produced by programmers and that using this information reduces the computational effort of diagnoses. Using information from the episodic learner model additionally speeds up the diagnostic process. The second study shows that ELM is able to predict individual solutions. Finally, correspondences and differences to related systems are discussed.