For example: Does the lead singer have a breathy voice? How distorted is the rhythm guitar? What key is it in? What tempo? Time signature? Are the lyrics about love? Sex? Loneliness? This comprehensive taxonomy makes it possible to use mathematical heuristic techniques to identify groups of similar songs. Pandora plays songs selected to match the various musicological characteristics of songs selected as inputs by the listener.
Pandora also allows the listener to give feedback on songs that play via "Thumbs up" and "Thumbs down". Thumbs indicate that the listener likes or dislikes one or more elements of that song. The Pandora algorithm responds by playing more songs that share characteristics with the positively rated songs – and fewer songs that sound like the negatively rated songs. Over time, each station will predominantly play songs the listener likes. Thousands of new songs are added to the collection every week. Those that share characteristics of songs positively rated by the listener will play too. Specific details on the inner workings of the MGP and Pandora algorithm's are mostly confidential.
Hope we continue to see you around the Community!
@AdamPandora , Fascinating writeup. I’d be curious to know what the approximately 450 attributes are. It would be cool if I could create a station based on, say, 12 of those attributes of my choosing. (12 is just a random number for this example.)
Probably outside your purview, but is there a reason that such a compilation must be proprietary?