If we think about the technology available around 2007, a crowd-sourced music recommendation system is basically what I would expect. No offense, that’s just what we had at the time. The Netflix prize wouldn’t come out until 2008. Pandora’s Music Genome project was perhaps ahead of its time but mostly manual (20-30 minutes per song!).
Today, we would usually use automatic song similarity technologies to recommend songs. The reason is pretty clear: popularity-based systems don’t know anything about the songs they recommend. They don’t know anything about you. They only know metadata about the song, that a certain number of people liked a song, essentially throwing away half of the information in a user-song tuple.
Today, we would usually use automatic song similarity technologies to recommend songs. The reason is pretty clear: popularity-based systems don’t know anything about the songs they recommend. They don’t know anything about you. They only know metadata about the song, that a certain number of people liked a song, essentially throwing away half of the information in a user-song tuple.