Skip to main content



A Fine-tuned Network

Every song written by every musician has at least one distinct feature that captures the attention of the listener and allows him or her to develop a strong connection or immediate dislike for it. Playing on this fundamental understanding, musicologists at Pandora strive to strip and analyze songs down to their technical characteristics, for instance the presence of multiple parallel 5ths or the use of pentatonic scales interspersed in the music backup vocals. However, the analysis is not limited to the aforementioned music theory technicalities. Amongst the numerous musical elements, musicologists also pay attention to a particular emotion the song evokes (and rates it on a scale of 1 to 5) or the dominance of a certain musical influence from a specific genre: swing music style from Jazz or skank guitar rhythm from Reggae music. As mentioned in the NY times article http://www.nytimes.com/2009/10/18/magazine/18Pandora-t.html?pagewanted=all&_r=0, “more than 700,000 songs, by 80,000 artists” have been scrutinized by this unique web of analysis. Also known as “The Music Genome Project”, it is the core technology developed by Will Glaser and Tim Westergren for Pandora.

Pandora aspires to make the user as the DJ – or the central focus of a highly connected network where the nodes are represented by the particular songs listened by the user and the edges are formed between the nodes based on the similarities shared between the songs. These edges can be formed and identified by a vast amount of similar musical traits between two songs, and these characteristics can range from the same genre the songs in comparison belong in, to the gender of the lead singers in the songs or the particular musical instruments used in the background music of both songs. A different number of networks can be formed with different nodes but with essentially the same edges. The nodes of the network can alternately be represented by the name of the bands and the edges between them can be identified as “strong” or “weak” based on the number of musical traits shared between the bands. These networks are all mapped out from the perspective of one particular user at a time. Pandora thus makes music recommendations with the importance of the user’s interest in mind instead of recommending music based on the popularity or number of “clicks” a music video link receives, a tactical trait used by YouTube. With a simple click of a “thumbs up” icon, Pandora introduces the user to a mapped musical world, drawing the listener into an extended and endless network of music and bands that are interconnected by shared characteristics.

Comments

Leave a Reply

Blogging Calendar

September 2014
M T W T F S S
« Aug   Oct »
1234567
891011121314
15161718192021
22232425262728
2930  

Archives