Well you could always regress on the blending coefficient, and let your data tell you whether it should be positive or negative. :)
On Tue, Apr 27, 2010 at 12:53 PM, Ted Dunning <[email protected]> wrote: > This is also very dangerous because negative ratings often correlated much > more tightly with what people like than with what they don't like. If the > blending is done carelessly, you can wind up recommending everything except > what the user likes (if the negative ratings overwhelm the positive in > spite > of the blending parameters) or you can wind up missing some very good > stuff. > > On Tue, Apr 27, 2010 at 12:46 PM, Jake Mannix <[email protected]> > wrote: > > > Another approach would be, if you have a lot of negative feedback as well > > as positive feedback, to train two models, both treating the feedback as > > positive, > > and then blend the results with a negative blending weight (i.e. subtract > > off a > > scaled multiple of the negative feedback recommendation score from the > > positive feedback recommendation score). > > > > This would only work if you had enough negative feedback to build an full > > model out of it. > > > > -jake > > > > On Tue, Apr 27, 2010 at 11:43 AM, Ted Dunning <[email protected]> > > wrote: > > > > > Boolean recommendations makes the assumption that either negative > > feedback > > > is really positive feedback (controversial, but works reasonably) or > that > > > negative feedback is not very useful (also controversial, but actually > > > generally true). > > > > > > If you really want to use negative information soundly, you need to use > a > > > much fancier algorithm. This can lead to huge costs in CPU and memory > > and > > > it may or may not help you. More practical is to use boolean > > > recommendations for the positive side and then just add a few > heuristics > > > for > > > the negative side. Typical heuristics include: > > > > > > - don't show items that a user has negatively rated (obviously) > > > - don't show items from a source that the user has rated negatively > > > several times without rating other items from that source positively > > > > > > > > > On Tue, Apr 27, 2010 at 10:27 AM, Tolga Oral <[email protected]> > > wrote: > > > > > > > We are building a system on top of provided recommendations allowing > > user > > > > to > > > > provide feedback. I have a question on how to apply the negative > > feedback > > > > for both boolean and none boolean scores. > > > > > > > > Assuming its a boolean item (item with no score), I cant see how we > can > > > > provide negative feedback for a given recommendation. Does this mean > we > > > > should convert our boolean items to a score of 1 and give score of -1 > > (or > > > > 0) > > > > for recommendations that user didnt like? > > > > > > > > This applies to items with score too I am not sure if the correct way > > of > > > > going about this is to provide 0 or a negative value. > > > > > > > > Thank you. > > > > > > > > > >
