[sangnoir]

> Even more bizarre is how they predict how many stars you'll give something, which seems fairly good. And then go ahead and recommend items with few predicted stars

To play the devils' advocate

* It's good for the user because their algorithm is not perfect and you might actually enjoy the 'few predicted stars' movie

* It's good for the algorithm because it allows fine-tuning in the cases of false-negatives and adds more data points

* having only high predicted stars in your list will probably move your Overton window on what's good/bad.

[rogerbinns]

There is no need for a devil's advocate argument - having some diversity in recommendations is good for many reasons. But they they need to start by getting the base recommendations done well, and then mixing in the diversity. Sadly they fail dismally at the former.