Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> Say you listen to Blue Lines by Massive Attack. Chances are, the recommendation engine will recommend you Portishead or Morcheeba.

Last.fm [1] recommends more Massive Attack, or "Aftermath" by Tricky, followed by Portishead. Which is fine, as you said "chances are" that's what's wanted. If you wanted UK rap, you at least used to be able to search for the tags "uk" and "rap", but I can't see that on the new website...

This service[2] uses the Last.fm API to make a Spotify playlist based on a track.

[1] http://www.last.fm/music/Massive+Attack/_/Blue+Lines

[2] http://www.spotalike.com/#playlist/97600



> If you wanted UK rap, you at least used to be able to search for the tags "uk" and "rap", but I can't see that on the new website...

In case of my example such a search would work. However, I guess there are lots of cases where a user doesn't even know what s/he wants. That is: "I want more of the same but it should also be very different"

Now that's quite a task for an algorithm! Also, making suggestions based on empirical user data is rather hard as people are highly emotional, especially when listening to music. Thus, people (in general) are not very predictable when it comes to music.

Individuals might have some core principles when choosing music so that perhaps individual algorithms for individual users could work but coming up with an acceptable one size fits all solution is rather impossible, imo.

Some people are happy when the algorithm suggests another top 40 song they haven't heard yet. Others will loose it when they get more metal songs but none of them is teutonic thrash metal [1]

Now we could empower people by making apps like spotify scriptable so that everyone could refine their recommendation algorithm. But realistically that is not what your average user wants...

Perhaps one could improve the situation by asking the user a few questions when s/he first starts the application. That way the app can choose the best fitting recommendation algorithm from a range of algorithms in the background.

[1] Apparently, this is a thing: https://en.wikipedia.org/wiki/Heavy_metal_subgenres#Teutonic...




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: