Student Strategies - Why did that student just click on a hint? To answer this question, you have to understand earlier student actions too. This is a very hard problem. We've been working on learning domain-independent models of student strategies. The biggest advantage of our approach is that it requires zero human-intervention. Though this often leads to difficult-to-interpret models, by using several education-oriented improvements, our new method actually produces models both highly predictive and easy to interpret. [early slides]

Thresholding for Variable Elimination - Most educational software uses complicated student and domain models that require many parameters. We explore the question of determining discrete threshold parameters such that variables previously predictive of learning can be shown to be non-causes.

Past Projects

Active vs. Passive Instruction - We conducted a lab study of tutorial options for Alice, a program for teaching computer science. The non-interactive tutorial showed significant advantages over the interactive tutorial. [paper]

Self-Explanation and Bottom-Out Hints - Deprecated by Student Strategies project.