5 ms·

Ask YC: feedback on my music recommendation site

7 points by maryrosecook 18y ago

http://theperceptron.com Enter the name of an artist you like and get back other artists you might also like. the perceptron's suggestions are based on actual people's music tastes: artists on the same label, artists who admire each other, artists who have played together, artists mentioned together in the press. Further, the site learns your music taste and uses what it knows to adapt its recommendations. I would love to hear the HN community's thoughts.

maxklein 18y ago
Seriously, WTF! I repeat, WTF! Your 'app' only has one feature, which is music recommendation, right? So it has no other features to make up for it. I go there and type 'Billie Holiday' and the first result that comes back is 'Iron Maiden'. Then comes a bunch of other seemingly random artists like Michael Jackson, Taj Mahal and so on. I don't see Ella, Louis, Anita or any of the people who usually belong to the Billie group. You can't release an application with one feature, and that one feature does not work. This is not reverso world where the opposite of working actually works. This is like a car being released that does not have a roof, and engine or tires. Fix your 'startup' and try again.
- furiouslol 18y ago
  well. he did follow the mantra of 'release early, release often' anyway, like what max said, if you release the app with only one feature, make sure that feature works. (eg. google - search, muxtape - make muxtapes) The recommendations spewed by your engine is pretty bad honestly. I don't see how a Michael Jackson fan would want to listen to Cyndi Lauper or Taj Mahal. Try bettering Pandora's recommendation engine. They're the ones to beat.
  - maxklein 18y ago
    Well, then my next startup will be: <html><head></head><body>
- maryrosecook 18y ago
  Yep, you've made a good point. Billie Holiday -> Iron Maiden != good recommendation. That sort of insanity will be filtered out as more music taste data is collected from users. Thanks very much for the feedback.
  - shafqat 18y ago
    Classy response after some pretty rough comments! Keep iterating and releasing. Good luck.
- davidw 18y ago
  I kind of like the idea of a Bille Holiday/Iron Maiden mashup. "Lady sings the UAAAAGGHGHGHGHGH"?
earle 18y ago
You clearly need to focus on a more algorithmic approach as the results you're returning are in no way better than a completely random selection from the same genre http://theperceptron.com/recommendation/artist/the+rolling+stones?h=true http://theperceptron.com/recommendation/artist/the+rolling+s...
- maryrosecook 18y ago
  Recommendations come from a number of sources, some good, some evidently not so good. However, the bad sources will get slowly suppressed as the site gets more user feedback. I'm working real hard on the algorithm right now to get the cream floating to the top sooner, and people's comments on HN have been really helpful. Thanks very much.

7 points by maryrosecook 18y ago

maxklein 18y ago

Seriously, WTF! I repeat, WTF! Your 'app' only has one feature, which is music recommendation, right? So it has no other features to make up for it. I go there and type 'Billie Holiday' and the first result that comes back is 'Iron Maiden'. Then comes a bunch of other seemingly random artists like Michael Jackson, Taj Mahal and so on. I don't see Ella, Louis, Anita or any of the people who usually belong to the Billie group. You can't release an application with one feature, and that one feature does not work. This is not reverso world where the opposite of working actually works. This is like a car being released that does not have a roof, and engine or tires. Fix your 'startup' and try again.

furiouslol 18y ago
well. he did follow the mantra of 'release early, release often' anyway, like what max said, if you release the app with only one feature, make sure that feature works. (eg. google - search, muxtape - make muxtapes) The recommendations spewed by your engine is pretty bad honestly. I don't see how a Michael Jackson fan would want to listen to Cyndi Lauper or Taj Mahal. Try bettering Pandora's recommendation engine. They're the ones to beat.
- maxklein 18y ago
  Well, then my next startup will be: <html><head></head><body>
maryrosecook 18y ago
Yep, you've made a good point. Billie Holiday -> Iron Maiden != good recommendation. That sort of insanity will be filtered out as more music taste data is collected from users. Thanks very much for the feedback.
- shafqat 18y ago
  Classy response after some pretty rough comments! Keep iterating and releasing. Good luck.
davidw 18y ago
I kind of like the idea of a Bille Holiday/Iron Maiden mashup. "Lady sings the UAAAAGGHGHGHGHGH"?

earle 18y ago

You clearly need to focus on a more algorithmic approach as the results you're returning are in no way better than a completely random selection from the same genre http://theperceptron.com/recommendation/artist/the+rolling+stones?h=true http://theperceptron.com/recommendation/artist/the+rolling+s...

maryrosecook 18y ago
Recommendations come from a number of sources, some good, some evidently not so good. However, the bad sources will get slowly suppressed as the site gets more user feedback. I'm working real hard on the algorithm right now to get the cream floating to the top sooner, and people's comments on HN have been really helpful. Thanks very much.