I believe that in the future that's the way to do it, but not right now.
Besides, I'm not a fan of automatic recommendations as they have a lot of drawbacks. Their major problem is: they won't help you find something you didn't know you liked.
Besides, I'm not a fan of automatic recommendations as they have a lot of drawbacks. Their major problem is: they won't help you find something you didn't know you liked.
This is patently untrue. It is more likely to find you stuff that you didn't know you like than the lowest common denominator strategy.
Just how exactly will they do it? I'm really interested, maybe I'm missing something?
Simply speaking, if a system knows all I listen to is classical music, it can recommend me a new classical composition, but it cannot all of a sudden recommend me dubstep (which I might actually like).
Regarding music taste, if the system knows that you have a strong correlation with a number of users who like classical but periodically explore new music, what they like will likely be something you might be interested in. This, in practice, is exactly how music discovery happens in the real world (that the discoveries of the people you share musical tastes with are more likely to interest you).
The system you are describing will find people who have a strong correlation with me and who also like:
1) Classical music and Justin Bieber.
2) Classical music and Metallica.
3) Classical music and Dubstep.
And it will recommend me Bieber, Metallica and Dubstep. I might like Dubstep, but I won't like Bieber and Metallica. How is it better than listening to random songs?