How best to discover?
Fred Wilson wrote yesterday about his preference for Last.fm over Pandora. His reasoning and others’ comments on the post reflect a sort of divide amongst online music consumers over what approach provides the best means for music discovery.
Last.fm is an internet music service that employs a “myware” application to track what you listen to on iTunes (or similar) and then (1) creates an internet radio station that reflects your preferences (similar to LAUNCHcast and MusicMatch’s old Radio MX service), and (2) points you to music “neighbors” (i.e. other users) on the network who share your preferences. You can then tune into your own personalized station – or those of neighbors – to discover new music. Collaborative filtering is applied to what you’re actually listening to vis-a-vis what others are listening to in order to suggest music and people that are a good “match” for you, presented in playlists streamed from a central database of content.
In constrast, Pandora – formerly known as Savage Beast and created through the Music Genome Project – employs people to analyze songs on a variety of musical attributes (e.g. melody, rhythm, instrumentation, etc). Listeners enter an artist name in a web-based application that will then play music that is similar to that from the noted artist, based on these assigned attributes. Songs are classified not on an overall basis (as one might do so with genres) but rather on a number of individual traits; taken together, these help guide a listener to music that is sonically similar to certain known music of interest, again presented in playlists streamed from a central database of content.
If one were to imagine a continuum with “social discovery” on the left side and “technological discovery” on the right, I would place Live365 (my current employer) on the far left, Last.fm in the middle, and Pandora on the far right. Live365 relies primarily on social discovery, although we use collaborative filtering to help guide listeners to other stations of likely interest. Last.fm uses collaborative filtering not only to guide a user to other people of interest but also to compile the playlists themselves (a user cannot actually pick the specific tracks and playlist ordering). On the other hand, Pandora omits a social component altogether and instead considers the inherent (though still subjectively-assessed) qualities of the music itself.
Ironically, technology is the starting point for Last.fm’s relatively social approach (i.e. the Audioscrobbler application) while people are the starting point for Pandora’s largely technological approach (i.e. musicians’ and others’ evaluations of music on a specific set of attributes).
I tend to lean towards the social approach myself (unsurprisingly). Friends and radio are the ways that people have traditionally discovered new music, and these means can be enhanced and combined online. On the internet, one can have access to “friends” from around the world (not just people known from physical-world interactions) and every person can create (or be represented by) a streamed, radio-style channel, enabling myriad channels rather than a handful constrained by scarce spectrum. While music analysis as offered by Pandora/Savage Beast and various others from years past (MoodLogic, Music Buddha, MongoMusic, etc) has its place, I think technologies that faciliate social connections (like collaborative filtering) are the most valuable.
Social discovery has a number of advantages over a purely technological approach:
— it allows consumers to more readily discover content that is not necessarily similar to what they currently like
— it transcends the intrinsic qualities of the music to pick up physical context, era, shared experiences, etc
— it potentially provides value beyond the music itself as can connect people with common tastes or interests
— it scales with (and, in fact, is generally enhanced by) growth in the amount of content and number of users
Filed under: discovery, ventures | 9 Comments