Enhancing a MOOC With Adaptive Learning

• By Linda L. Briggs
• 05/14/14

When Ohio State University math professor Jim Fowler and his colleague Thomas Evans, a senior instructional designer at the school, decided to create a calculus MOOC, they envisioned a course with an adaptive learning engine that would give students lots of practice doing math problems that matched their skill level. But when the pair moved forward with building the course in Coursera, they came up with a problem: The system did not include tools to support their adaptive learning agenda.

Coursera's mechanism for delivering problem sets is designed to provide the same number of questions to every student. Fowler and Evans envisioned something different: "I really like being able to have the number and sort of questions students do depend on their performance," Fowler said. "If a student is really just killing it, they can move through the whole thing quickly. If students are struggling, they get more time to think about it. I think it’s a way of being more respectful of a student’s time."

To give their Coursera course the adaptive flavor they were looking for, Fowler and Evans created an add-on, dubbed "MOOCulus." Bolted on to Coursera’s MOOC platform, the tool is designed to feed students progressively harder questions based on previous answers.

"We felt it was important that students have this sort of practice area," Evans said. "Learning calculus involves doing math, lots of math.... We wanted problems that would react to students."

MOOCulus is written in JavaScript and the open source Web application framework Ruby on Rails. Although geared toward calculus, the tool is freely available for others to tailor to their courses. Fowler did the bulk of the backend programming, along with some graduate students; Evans handled other aspects, including production issues and interacting with Coursera.

How the Course Works

The calculus course, which launched January 2013, is an interesting hybrid, residing partly on the Coursera MOOC platform and partly outside it. The adaptive learning tool itself sits at MOOCulus.osu.edu and runs external to Coursera on OSU servers, but students seamlessly log in using Coursera credentials. Video lectures from the course are also freely available on YouTube. Additional quizzes and content reside in Coursera, and certificates are awarded only for coursework done inside Coursera.

Since the course was first offered, it’s been expanded and is now available on Apple’s iTunes U, Evans noted. The self-paced model there allows students to move through the course at their own speed.

Fine-Tuning Adaptivity

A huge challenge in building the adaptivity of MOOCulus, Fowler said, was accurately measuring student understanding. As a student works through a problem, hints are available. The software weighs the hints used and the amount of time taken to answer the question, as well as the student’s answer -- whether right or wrong -- to determine which question to display next.

The MOOCulus software also collects participation data in the background, and the level of understanding of the current concept is displayed to the student on a color-coded progress bar that inches along from red to green, indicating mastery. Unlike a paper quiz, where each student, regardless of level, gets the same number of questions, MOOCulus "keeps on providing activities that are at the appropriate level," Evans said, thus enticing students to do more problems.

But forum feedback from participants illustrated just how challenging it is for an adaptive engine to serve just the right amount of an exercise, at the right level, to each student. Parameters in the adaptive backend code sometimes served extremely difficult exercises, and too many of them, Fowler said, discouraging some students. On the other hand, some exercises may have been too easy, and Fowler continues fine-tuning the adaptive engine.

Beyond adjusting the problem sets, the MOOCulus engine also provides immediate, data-driven feedback on the effectiveness of online instruction, pointed out Fowler and Evans -- information that would take an impossible number of years of teaching to gather otherwise. As students answer questions, MOOCulus can collect vast amounts of data on learning patterns. In just a 10-week course, for instance, Fowler’s students spent the equivalent of 10 "person years" solving problems in MOOCulus, delivering two million correct answers. That data, in turn, shows which questions are too easy or too hard, or which might be better predictors of a student’s overall performance in the course. The data can then be used to create better classes, questions, tests and instruction. There are other potential uses as well; for example, since the adaptive engine measures how long a student is spending on an activity, an instructor or graduate student could theoretically jump in when a student is clearly confused.

Student reaction to the course has been highly positive, especially with regard to the adaptive learning engine. "People definitely like having a lot of exercises to do," Fowler said. In a subsequent Calculus 2 MOOC he taught, "we tried to do everything in the Coursera quizzes, and there were comments in the forum about people missing having the [MOOCulus] exercises."

Fowler added that the course has led to some interesting conversations on pedagogy. "Suddenly it’s OK for people to have a lot of teaching conversations that I wouldn’t necessarily have had before, and that’s taught me a lot," he said. Those conversations have included what makes an effective classroom experience, what the ideal classroom looks like, and how to keep more students engaged and simply doing more math.