On Mon, May 6, 2013 at 12:50 PM, Koobas <[email protected]> wrote: > Since Dominik mentioned item-based and ALS, let me throw in a question > here. > I believe that one of the Netflix price solutions combined KNN and ALS. > > 1) What is the best way to combine the results of both? >
I think that combinations are important, but I think that the combination of very similar kinds of algorithms working on essentially the same data has almost no practical impact. 2) Is there really merit to this approach? > Yes, but. > 3) Are there other combinations that make sense? > (user-based + item-based)? > Absolutely. But I really think that the real mileage for improvement comes from the following: a) combining different kinds of behavior into a single recommendation framework b) judicious use of dithering to improve exploration c) substantial UI improvements to gather additional exploratory data for the recommendation engine d) principled testing framework for design and algorithmic alternatives Minor algorithmic changes are almost not visible on the priority list. Once you hit "pretty-good" there are far more important things to take on in a real rec engine project.
