3 ms·

I am not sure whether this is a problem that can be solved entirely with software. Oftentimes it is not so obvious what's the relationship between what you have

by cJ0th 11y ago

I am not sure whether this is a problem that can be solved entirely with software. Oftentimes it is not so obvious what's the relationship between what you have and what you'd like to discover. In fact, one of the biggest issues I have with current methods is that they recommend me more of the same. When I look for new music I want - to some extend - something that is unlike what I've heard before. Say you listen to Blue Lines by Massive Attack. Chances are, the recommendation engine will recommend you Portishead or Morcheeba. (User likes genre trip hop? let's play more trip hop). But maybe I'd like to explore different forms of UK rap music now...

Symbiote 11y ago
> 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 http://www.last.fm/music/Massive+Attack/_/Blue+Lines [2] http://www.spotalike.com/#playlist/97600 http://www.spotalike.com/#playlist/97600
- cJ0th 11y ago
  > 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_thrash_metal https://en.wikipedia.org/wiki/Heavy_metal_subgenres#Teutonic...

by cJ0th 11y ago

Symbiote 11y ago

> 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 http://www.last.fm/music/Massive+Attack/_/Blue+Lines [2] http://www.spotalike.com/#playlist/97600 http://www.spotalike.com/#playlist/97600

cJ0th 11y ago
> 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_thrash_metal https://en.wikipedia.org/wiki/Heavy_metal_subgenres#Teutonic...