On Sun, Jul 3, 2011 at 1:08 PM, Sean Owen <[email protected]> wrote: > I don't see why one would believe that the randomly selected items > farther down the list are more likely to engage a user. If anything, > the recommender says they are less likely to be engaging. >
There are two issues with this logic. First, on a second visit, items that the user has already decided not to click should have some penalty. Secondly, the ordering is only what the recommender says given the data so far. Getting exploratory data may compromise the click-through on *this* presentation, but it makes later presentations, possibly to other users better. (Or put another way, by this reasoning, we ought to pick > recommendations at random.) > Dithering isn't a uniformly random presentation. It starts with the recommender's list and broadens it a bit. > I do think that it's possible that a recommender isn't quite capturing > the utility of recommendations correctly. But then that's an issue of > the algorithm. > It is also an issue of narrow training. Getting broader training data helps (a lot). > > On Sun, Jul 3, 2011 at 9:02 PM, Konstantin Shmakov <[email protected]> > wrote: > > It seems that as long as recommenders are dealing with the "economy of > spam" > > (most users are not interested) any additional engagement e.g. through > > randomization or more reach recommendations would help. Is that right? >
