Transcript Course Helper: A Course Recommendation System

Course Helper: A Course
Recommendation System
Fred Wulff
The Problem
• Axess stinks
• We’d like to do better - do the whole
Amazon thing, and help people find and
discover
• Couple of issues interesting from a data
mining perspective
Data Extraction
• Axess
• After spidering it, you end up with
“interesting” web pages
• “Late adolescent” algorithm
• Capitalize on the tendency of English
speakers to organize left-to-right, top-tobottom
Axess
Recommendations
• Traditional document similarity measures
don’t work that well
• Suggests collaborative approach
• However, problems!
• Mixture: Item-item based on Facebook data.
Add points based on Facebook links
(normalize) and combine with major data.
• O(n*k) where n is the number of courses in
history and k is the average branching factor
(30 in practice for the data I was using)
• 70ish% success (loosely defined)