4 ms·
The BellKor solution to the Netflix Prize
- aston 19y agoInteresting that the percentage difference between the winning solution with 107 blended approaches and the close-by solutions using only one approach was on the order of a fraction of a percent in improvement.
- timr 19y agoYet another real-life application of the 80/20 rule.... In computational biology (my world), it's routine for an initial study to account for most of the eventual success of a method, with many man-years of work dedicated to improving success rates only by tiny fractions. (As an aside, if you're an ambitious researcher, the implication of this is clear: don't waste your time refining things. Rapid testing of ideas is more important than squeezing out that last 1% of performance improvement. Iterative refinement is research death.)
- ntoshev 19y agoDo you have a reference describing the single approach?
- aston 19y agoIf you click around the links on the leaderboard [http://www.netflixprize.com/leaderboard http://www.netflixprize.com/leaderboard], a lot of folks describe the basics of their approaches. There are also a few entries (like "When Gravity and Dinosaurs Unite") that are explicit blends between two single approaches from different teams.