| Intelligent tutoring systems can provide benefits of one-on-one instruction automatically and cost effectively. To make the intelligent tutoring systems as effective as expert human tutors, the utility of a computational model of expert tutoring in generating effective natural language feedback in intelligent tutoring systems is demonstrated.;To set up a basis for computationally modeling expert tutoring, a comprehensive study of the difference between one expert tutor and two non-expert tutors in effectiveness, behavior and language is presented. Based on the empirical results, a rule-based model of expert tutoring is developed, which takes advantage of a machine learning technique, Classification based on Associations. To employ the model of expert tutoring in the natural language feedback generation for intelligent tutoring systems, a framework of feedback generation with 3-tier probabilistic planning is designed. The 3-tier planning automatically generates, selects and monitors plans for generating effective tutorial feedback based on the rule-based model and the information state which keeps track of the interaction in the intelligent tutoring system. The evaluation results show that the tutorial rules successfully model expert tutoring and the intelligent tutoring system using them improves learning significantly. |
