Machine Learning and Friends Lunch
Applications of machine learning to education: Inferring learning and attitudes from students' behavior in a tutoring system
Ivon Arroyo
UMass
Abstract
Intelligent tutoring systems are pieces of software that teach a
student about a domain, and change what and how they teach based on
some internal model of the student's knowledge and affect (like a
good teacher would). In the past, these student models have been
built based on cognitive theories of learning. However, student
models may be built (or enhanced) using machine learning and
statistical techniques over data from past users.
My contribution is the idea that very generic behaviors of a student
interacting with the system may reflect more abstract users' goals,
attitudes and amount of learning with the tutoring system. These
behaviors may be summarized and retrieved from user log files. I
present and evaluate the accuracy of a Bayesian Network learned from
data that allows to infer students' learning, attitudes towards
learning and perceptions of the system from behaviors related to
problem-solving time, mistakes and help requests.
The final goal is to make tutoring software that dynamically adapts
to students' affective and intellectual needs, in order to maximize
the educational experience.