Using Association Rules to Guide a Search for Best Fitting Transfer Models of Student Learning
Jonathan E Freyberger · 2004
Abstract: We say a transfer model is a mapping between the questions in an intelligent tutoring system and the knowledge components (i.e., skills, strategies, declarative knowledge, etc) needed to answer a question correctly. [JKT00] showed how you could take advantage of 1) the Power Law Of Learning, 2) an existing transfer model, and 3) set of tutorial log files, to learn a function (using logistic regression) that will predict when a student will get a question correct. In the main conference proceeding [CHK2004] give an example of using this technique for transfer model selection. Koedinger and Junker [KJ99] also conceptualized a search space where each state is a new transfer model. The operators in this search space split, add or merge knowledge components based upon factors that are tagged to questions. Koedinger and Junker called this method learning factors analysis emphasizing that this method can be used to study learning. Unfortunately, the search space is huge and searching for good fitting transfer models is exponential. The main goal of this paper is show a technique that will make searching for transfer models more efficient. Our procedure implements a search method using association rules as a means of guiding the search. The association rules are mined from a dataset derived from student-tutor interaction logs. The association rules found in the mining process determine what operations to perform on the current transfer model. We report on the speed up achieved. Being able to find good transfer models quicker will help intelligent tutor system builders as well as cognitive science researchers better assess what makes certain problems hard and other problems easy for students.