Stamper, J. (2010). AUTOMATIC GENERATION OF INTELLIGENT TUTORING CAPABILITIES VIA EDUCATIONAL DATA MINING. Unc Charlotte Electronic Theses And Dissertations.
Intelligent Tutoring Systems (ITSs) that adapt to an individual student's needs have shown significant improvement in achievement over non-adaptive instruction. This improvement occurs due to the individualized instruction and feedback that an ITS provides. In order to achieve the benefits that ITSs provide, we must find a way to simplify their creation. Therefore, we have created methods that can use data to automatically generate hints to adapt computer-aided instruction to help individual students. Our MDP method uses data from past student attempts on given problem to generate a graph of likely paths students take to solve a problem. These graphs can be used by educators to clearly understand how students are solving the problem or to provide hints for new students working the problem by pointing them down a successful path to solve the problem. We introduce the Hint Factory which is an implementation of the MDP method in an actual tutor used to solve logic proofs. We show that the Hint Factory can successfully help students solve more problems and show that students with access to hints are more likely to attempt harder problems than those without hints. In addition, we have enhanced the MDP method by creating a "utility" function that allows MDPs to be created when the problem solution may not be labeled. We show that this utility function performs as well as the traditional MDP method for our logic problems. We also created a Bayesian Knowledge Base to combine the information from multiple MDPs into a single corpus that will allow the Hint Factory to provide hints on new problems where no student data exists. Finally, we applied the MDP method to create models for other domains, including Stoichiometry and Algebra. This work shows that it is possible to use data to create ITS capabilities, primarily hint generation, automatically in ways that can help students solve more and more difficult problems, and builds a foundation for effective visualization and exploration of student work for both teachers and researchers.
