The Netflix Effect: When Software Suggests Students' Courses

One of my professors just mentioned the other day that some CS grad students built a course suggestion engine. It apparently worked - all they needed was the cooperation of the registrar's office to incorporate students' course history.<p>The registrar refused.

Even without knowledge of previous performance, computer-assisted course selection can be great.<p>A friend of mine implemented a genetic algorithm (in Javascript) that mutated possible timetables until it seemingly found the best combination of your chosen criteria. You could select things like "I like Professor X", "I don't want to have lectures on wednesday", "I don't want to get up before 9" and give the algorithm incentives/disincentives for specific lectures.<p>Worked remarkably well.

Course suggestion engines are pretty easy to start with due to limited choices and mandatory paths in programs. In fact I implemented something akin to this as part of a university assignment.<p>The tool was actually meant to be provided for course coordinators to estimate number of students who are likely to take the course in next semester and sometimes as far as the length of a normal program.<p>Post-release, maybe past peer history could be used for accuracy. It would not solve the volatile nature of new courses though.

That would be awesome to have in our school.<p>All the apps people are making here seem to be confined to course scheduling (whether something conflicts, etc.) or courses we need to graduate. They haven't actually gone into the realm of recommending possible courses we'd be willing to take.<p>As the article mentions, the schools have all these data they're not using. They can probably predict which students would succeed in courses (they have access to grades), know which similar courses students may be interested in (cross-reference with previous courses taken and see what others have taken with a similar plate) and even "learn" the scheduling habits of the student (see the previous class times the person usually chose for clases - i.e. a morning person scheduling all classes in the morning or someone scheduling all classes at night).<p>As long as we still have a choice (to accept or reject the recommendation), something like this would be useful (another resource to inform our decisions).

Computer-generated suggestions seem likely to me to be "the future". They're getting to be easy to implement, so I doubt it'll be long before they're literally everywhere, and we get used to computers being creepily accurate at predicting what we want.

Interesting to see others doing this. I'm working on a similar project at Brown right now with a few friends. We're lucky - the Registrar's office agreed to give us anonymized data to bootstrap our recommendation system.

This should be interesting, current course ladders right now are too rigid or very limiting to exactly curate a student's academic career. While it has it its downsides, this would be an experiment worth watching.