Learning Analytics via Sparse Factor Analysis

Introduction Textbooks, lectures, and homework assignments were the answer to the main educational challenges of the 19th century, but are now the main bottleneck of the 21st century. In particular, today’s textbooks are typically static, linear in organization, time-consuming to develop, soon out-of-date, and expensive. Lectures remain a primarily passive experience of copying down what an instructor says. Homework assignments that are not graded for weeks provide poor feedback to students on their learning progress. Even more importantly, today’s courses provide only a “one-size-fits-all” learning experience that does not cater to the background, interests, and goals of individual students. In contrast, we envision a statistically-minded, machine-learning based, cognitive tutor that is able to learn about the student as the student learns about the subject material being taught. This approach would allow the cognitive tutor to naturally assess which knowledge areas the student understands well, as well as the areas that remain problematic, enabling crucial tasks such as the automatic recommendation of remedial study material or additional practice problems [1].