Representing a student’s learning states and transitions

We describe an ongoing project to develop an adaptive training system (ATS) that dynamically models a student’s learning processes and can provide specialized tutoring adapted to a student’s knowledge state and learning style. The student modeling component of the ATS, ML-Modeler, uses machine learning (ML) techniques to emulate the student’s novice-toexpert transition. ML-Modeler infers which learning methods the student has used to reach the current knowledge state by comparing the student’s solution trace to an expert solution and generating plausible hypotheses about what misconceptions and errors the student has made. A case-based approach is used to generate hypotheses through incorrectly applying analogy, overgeneralization, and overspecialization. The student and expert models use a network-based representation that includes abstract concepts and relationships as well as strategies for problem solving. Fuzzy methods are used to represent the uncertainty in the student model. This paper describes the design of the ATS and ML-Modeler, and gives a detailed example of how the system would model and tutor the student in a typical session. The domain we use for this example is high-school level chemistry.