Machine Learning in Online Environments

Machine learning techniques (including such things as Bayesian knowledge tracing and its successors) are a big deal in online education these days. You might not see it implemented right now, but unless things go very wrong, you’ll see it in the next five years. It will definitely show up behind the scenes; hopefully it will also show up as a major explicit feature.

The typical goal is personalization of learning, with the assumption (which seems fairly warranted to me) that such personalization can improve the learning process.

If you aren’t quite sure what could be done with machine learning and personalization in a course, let me give a few examples.

  • If Student S got the last three problems of this type right on the first attempt, don’t give S any more of this type of problem for a week.
  • If S hasn’t seen a problem of a particular type for three weeks, add one to the next assignment. If they get it right, mark them for mastery in that topic.
  • If S has mastery of a topic for which the current topic is considered a prerequisite, give S half the number of problems of this topic. If they get some wrong, give them the full set of problems and weaken the “prerequisite” connection between the topics.
  • If the average score on assignment A1 is too low, add assignment A2 to the course.
  • Randomize the order of resources R1 through R4 and see whether one particular order works better in terms of later performance.
  • Feed each student problems P1 through P5, but randomize who gets P6 through P10 so we can see which ones are most effective at helping students learn. Then fix up their grades afterwards so we aren’t costing anyone points.
  • If the successful guess rate is high for a problem, remove it.
  • Find out which categories of students are helped most by which sorts of exercises.
  • Adjust the wording of your resources to match the reading level of your students, or to push them toward higher reading levels.

All of these are fairly simple rules. You can work with individual students, with groups, with sections, with whole classes or even cohorts. You can recommend problems, book chapters, group member changes, even tell people which courses they’re likely to pass. Hell, some schools do that already.

The major barrier to including things like these in existing online education systems is that most people don’t know about them yet. Education researchers generally do, and people in AI research, but not the folks who program online education systems. There’s also a “just get it working” ethic that is fairly common in online education right now. That makes this a second- or third-generation technology – but a hell of a promising one.


About Colin Fredericks

By day I help to create online courses at HarvardX. By night I write roleplaying games.

Posted on September 25, 2013, in Uncategorized. Bookmark the permalink. Leave a comment.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: