NEURAL NETS FOR INTELLIGENT TUTORING SYSTEMS

Intelligent tutoring systems are real-time adaptive systems aimed at tailoring the teaching strategy to the current state of the student. An essential prerequisite of these systems is therefore the capability to model the cognitive state of a specific user. While present day models in ITS's are based upon symbolic AI, in this paper we describe a neural approach to the problem. The network learns to extract the relevant part of the student's behaviour by analyzing his problem-solving strategy on a set of exercises. An example in the field of high-school physics is discussed. The major results are that i) the network is capable of learning different cognitive styles with high precision and ii) the network can handle and even diagnose inconsistencies in the student's behaviour.