The original algorithm uses a mix of strategies to select songs. Its more or less hand picked. One thing that makes Grooveshark special is that it is all user-submitted content which is terrific from a selection and discovery point of view but an unwieldly problem from a data point of view. Note that our library stands somewhere around 6 million songs compared to Pandora's 600,000 so generating recs has to be done at great scale. The way in which users interact with our site is fundamentally different as well because you can add any mix of songs to your queue at any period in time.