UMass Machine Learning and Friends Lunch | Main / Educational Data Mining Learning To Teach From Student Logs Of A Tutoring System

Educational Data Mining is a new area of research (educationaldatamining.org) concerned with developing methods for exploring the unique types of data that come from educational settings, and using those methods to better understand students and create better educational software. In this talk, I will present the work towards a general methodology to make tutoring software "smarter at teaching" from past student logs. I model student effort at individual practice items in a mathematics tutoring software from a collection of student behaviors (e.g. timing, accuracy, and help requests). This integral and varied view of independent behaviors at practice activities helps to discern factors that affect student behavior beyond cognition (e.g., help misuse due to meta-cognitive or affective issues). Several tests at different levels guarantee that estimates are not biased to the variety of mechanisms in place at the moment of data collection.

This work was honored with the best paper award, and presented at the 3rd International Conference in Educational Data Mining, held at CMU, Pittsburgh, PA, this past summer.