| A model for intelligent tutoring is discussed whose aim is to provide an adaptive system that directs a student to high-level mastery of learned concepts. Several open problems exist in intelligent tutoring, including adaptivity, sufficient subject matter, explanations, student modeling and identification and revision of beliefs. Expert-system-based tutors lack sufficient subject matter while hypermedia-based tutors lack inference knowledge and cannot provide intelligent tutoring. We have combined these two technologies for detailed study material as well as intelligent interaction with a student. Our model components include the Tutoring Strategy Module, Student Modeling Module, Expert Problem-Solver, Knowledge-base, Dynamic Questions Module, Study-Management Module, and the hypermedia component providing an electronic textbook and a browser. The adaptive and opportunistic tutor addresses misconceptions as they occur, and receives adaptation input from the Modeler. We also propose the Collection and Invitation intervention technique by which the tutor would not offer solution assistance to a student unless otherwise directed. Our Modeler differs from the traditional overlay and buggy models by performing solution-strategy-independent modeling based on the correctness and completeness of a student's solution. Using the final result obtained by the Expert Problem-Solver, it uses the knowledge-base to compute a set of paths to the same solution, identifies student beliefs from interactions, and identifies which paths are applied by the student based on the frequencies of beliefs in the computed strategy-base. The Modeler does not employ a mal-rule-base but labels beliefs and identifies misconceptions and missing beliefs by processing the knowledge-base. It also performs Instructional Belief Update by maintaining two belief sets and removing contradictions from an opposing belief-set before arriving beliefs are stored. For belief revision we have defined an improved coherencist scheme based on the resolvent model developed by Dalal (1988). We have also specified algorithms for components of the Modeler which are of reasonable running times. The Expert Problem-Solver as well as the alternative-path-computation phase of the Modeler have been implemented.;The thesis has contributed to intelligent tutoring research in three ways: solution-strategy-independent modeling, identification of misconceptions using the knowledge base rather than a mal-rule set, and the improved belief revision scheme which incorporates derived beliefs, adjusts the status of persistent beliefs whose justifiers have been detected, and does not fail in multiple contradictions. |
