Can we predict which groups of questions students will learn from?

In a previous study ([4]), we used the ASSISTment system to track student knowledge longitudinally over the course of a schools year, based upon each student using our system about a dozen times during the course of the year. This result confounded learning from the computer system with students learning from their sitting their normal class. In this work, we look to see if students were reliably learning from their time spent with the computer in a single day. Our result suggests that students performed better later in the same computer session on similar skills, which indicates students are learning from using ASSISTments. However, learning is rather uneven across groups of skills. We test a variety of hypotheses to explain this phenomenon and found that the automated approaches we tried were unable to account for the variation. However, human expert judgments were predictive as to which groups of skills were learnable.

• Type: Conference
• Year: 2008
• Where: EDM
• Authors: Mingyu Feng, Neil T. Heffernan, Joseph E. Beck, Kenneth R. Koedinger
• Keywords: Data Mining, Dozen Times, EDM 2008, Schools Year, Student Knowledge