Name: The Music Genome Project and Pandora
URL: Pandora.com
Category: Music Taxonomy/Metadata/Music Streaming
Location: US
Why is this of Interest:
Pandora and Last.fm make a fascinating comparative study of relational database design in a network environment. We can think the difference between these two engines as a choice between a topological or typological taxonomy. Thay are both ‘dynamic media’ distribution engines that develop and deploy different kinds of ‘fringe intelligence’ or ‘relational dyanmism’ in order to recursivley populate their respective DB’s.
Brief Description:
The Music Genome Project is a commercial project aimed at developing a database of music information classified according to a defined taxonomy by paid musicians who do the analysis manually. Pandora is a streaming engine that uses this database to link like artists into a user’s playlist. The streamer operates within a browser (flash interface) and has options for labeling a selected track as inappropriate. A user can build radio playlists based on a selction of artists or a single artist. Recent additions have included the ability to search other shared stations adding a degree of social interaction to the streaming side of the project. This data does not feed back into the database which is essentially an ‘expert’ system. There are serious limitations on the user interaction due to radio licensing constraints (6 ‘skips’ an hour) – Only US registration is allowed although easily evaded because there is no check of the user IP.
Development History:
Development began with the Music Genome Project in early 2000. Tim Westergren is the founder and CEO and appears fundamental to the projects ongoing development. The Pandora Streamer was added as a subsidiary to the project but has become a central component in attracting their own user-base. MSN (Microsoft Network) uses Pandora profiles to power its play-listing. The future of Pandora is apparently threatened with a change to web radio licensing that will see the costs of streaming copyrighted material threefold. See the accompanying interview link for more information.
Analysis/Comparison:
Pandora and Last.fm are often considered competitors offering similar functionality and discussion regarding both engines tends to be reduced to which one provides a better listening experience. Few commentators appear to understand the difference between either the functionality or the philosophies of these two projects. I should declare that most users choose one or the other and stick with it and I am a last.fm devotee. Last.fm takes considerable time and input to develop a consistently appropriate playlist and this often means users preference Pandora’s rather simpler ‘user-directed’ approach/experience. Last.fm relies on scrobbling a users listening history and its value is augmented by the diversity of a user’s music library. The listening experience on Pandora is rather more regulated according to the Artists added by the user to their initial choices. This means that Pandora recommendations are based on the classification system, according to their so-called ‘genome’ which is prescribed by staff according to a proprietary schema consisting of over 400 ‘qualities’. Playlists tend to be successful in transcending social contexts to suggest like music regardless of the history, trends, popularity etc. This approach effectively circumvents the traditional recording industry/marketing tilted focus of broadcast radio; 14 year olds might now realize that Wolfmother has more in common with Black Sabbath than they do Eskimo Joe and Jet has more in common with The Kinks than..well..any of their contemporaries. Playlists are also very successful in finding like artists that are otherwise marginalized in a particular genre because they didn’t achieve international distribution or simply didn’t get the marketing support that others did. In my opinion Pandora suffers because its playlist tend to be rather predictable but many find Last.fm far too random for comfortable listening. On last.fm I am likely to have Merzbow followed by Monolake followed by Coltrane while on Pandora the playlists tend to be more consistent due to the Schema and its status as an ‘expert’ system.
Pandora and Last.fm make a fascinating comparative study of relational database design in a network environment. We can think the difference between these two engines as a choice between a topological or typological taxonomy. The Pandora model suffers the traditional problem of having a relatively small slowly developing expert edited DB but on the other hand it tends to avoid the questionable ‘wisdom of the crowd’ and delivers a consistent and efficient recommendations engine. Both engines, Pandora and Last.fm, suffer by varying degrees because they rely in the users input in the first instance; Pandora is unlikely to show you those new genre’s that you didn’t know you were looking for while Last.fm displays a tendency to catch mainstream tastes in mainstream circles. This latter point should alert to the value of a degree of randomness (the elicited yet involuntary balance again) in any attempt to create a generative relational database system.
Other links:
Tim Westergren Interview @ Create Digital Music : An extensive Interview about Pandora and the Music Genome Project.
Follow up about the RIAA changes to webcasting license costs