Basically: record every song you listen to, and when you listen to it (plus some more metadata), and then add it to a giant public dataset. Open source software then uses that to make music recommendations based on your and other people’s listening, and to give you interesting stats about your listening.