Courserank: a socially-networked course selection system for Stanford
Georgia Koutrika et al. from Stanford built a system called “Courserank”—a place to evaluate courses. It has been used by 98% of freshman at Stanford. In addition to just picking courses, it offers a number of interesting social features. It’s planner page lets students enter their entire plan for all years in school, classes and when you plan to take them. Then after the term they enter star ratings and grades. There’s also a tab showing which students plan to take which courses in which term, so you can take with students you know. A Requirements feature that tells you what you still need to take. There’s the opportunity to enter reviews and have discussions about various classes. Students actually voluntarily enter their grades so you can find out grade distributions in the class.
Courserank is a social site. But some things set it apart. It’s not open, but has a well defined closed community—only Stanford ID can participate. It’s not flat—it has well defined distinct constituencies (undergrads, grads students, faculty, the school). It has special purpose tools like course planner that ar e highly domain specific. It makes hybrid of user data and “official” data. These make it a “special purpose social site”.
They wanted grade distribution for courses, but didn’t have access to official records. So relied on students entering their own grades. They compared the results to some official data: they follow each other very well. People are honest, unlike big social sites. Perhaps because this is a closed community? Or because the tool is actually helpful to them so they are motivated to “give back”?
They then swiched to a different speaker who spent time talking about recommendation systems. They want to support general recommendation. What courses should I take based on my background? What major should I pick based on my performance? What is the best semester to take AI? Current recommendation systems are not flexible enough or extensible to allow these kinds of questions. They’re trying to create a formal model of recommender questions, kind of an extensions of database queries with a “recommend” operator.