Notes from the “Help! My iPod Thinks I’m Emo!” panel
Here are some notes from my part of the talk. Paul Lamere may share some soon too, he naturally disagrees :)
WHAT IF THERE IS NO DISCOVERY ALGORITHM?
Current music recommendation systems have tons of serious issues: they don’t properly incorporate long tail content, fail in offering anything close to serendipity, and get even more confused by divisive content (see the ratings for Napoleon Dynamite on Amazon for an example). Smart people with good funding have been working to solve all of these issues, but the progress seems limited at best. What if we are working to solve the wrong problem?
WHERE ARE WE TODAY?
Pandora’s popular music characteristic-based method offers listeners a “give me more of the same” experience. Last.fm’s social listening analysis leads you from one popular act to another with limited opportunity to stumble into the unknown as a by-product of attempting to filter noise and satisfy the greatest number of people possible with recommendations. Echonest’s hybrid recommender incorporates audio analysis, blog content and traditional collaborative filtering techniques for a seemingly promising mix, but the results have yet to be fully available in a significant consumer offering. These approaches are valuable if audience size (Pandora’s millions of listeners) and business acquisition prices ($280M for Last.fm) are considered, but their cultural impact & influence pales in comparison with destinations producing editorial, such as Pitchfork, select music magazines and music blogs.
TASTE IS IRRATIONAL
These issues exist because the ultimate nature of human taste is irrational and depends on factors impossible to capture with computer systems. Interpersonal relationships, social pressures, locales, context in which music is encountered, meaning that individuals assign to a music encounter (has a girl you liked ever introduced you to music you just happened to also like?) all make a huge difference in how people respond. Much of this happens outside of the clean social graph of Facebook or the neatly recorded playback logs of music sites. All of these factors affect each individual differently at various stages of their lives, further complicating things.
ACCEPT IRRATIONALITY
The way to drive genuine discoveries without a significant reliance on collaborative filtering or recommendation algorithms is to intelligently select and present information that captures the context of a particular piece of music and creates meaning for the person interacting with the system. Use computer systems to connect people, spotlight individual voices, then have voices and social connections define what music everyone interacts with.
EXAMPLES
I am not the first one to suggest this approach, and many of these elements have been successfully implemented by a variety of music sites on the web. Some of the sites below also mix in collaborative filtering, but the real value in those environments is still produced by context and meaning that is created and presented.
Ishkur’s Guide to Electronic Music - Unique presentation and editorial
Written in witty style and presented in an esoteric flash environment, Ishkur’s Guide capitalizes on mixing playfulness with an authoritative voice.
Guitar Hero, Rock Band Gameplay - Context & Interaction
These games have created some of the most unique and entertaining interactive music experiences to date. The very first encounter many people will have with a whole set of artists will now be via these games.
imeem.com, MySpace Music Social Playlists - Context
What’s on the playlists of your friends? It’s important to consider the quality of execution as well as the presence of your real friends on any of these networks. If they are there, however, it’s hugely valuable.
14 Tracks: Weekly Themed Track Selections - Context & Meaning via Editorial
A weekly themed selection of tracks sold at a discount (50% off) at the Boomkat store. The themes dive deep into specific genres or are centered around a single idea like “14 Tracks That Make Me Wish I Played Piano.” This arrangement makes buying and discovery compelling by the careful curation while making the impulse buy easier.
thesixtyone - Playful Presentation & Meaning
A game-like approach to music discovery. Users (or players?) are encouraged to sample more obscure songs to earn points they can later use to earn cool privileges on the site. The meaning is created by the invented value and abilities that the points provide.
The Hype Machine - Visible Context & Meaning
What are music-obsessed fans talking about? Music blogs written by music writers and passionate fans alike provide a window into what’s awesome and interesting in music at this moment. The selection of the blogs keeps the quality up and reduces noise and lower common denominator issues present in collaborative filtering
Pitchfork - Meaning via Authoritative Editorial
The best known site for independent music. Since 1995 (3+ web startup lifecycles), Pitchfork has created a strong brand of opinion that is often divisive but never ignored.
WHAT’S NEXT?
Mechanical music recommendation will become an even bigger commodity. Recommending users or voices, on the other hand, will still have significant value.
Big jumps in value-creation will come either from re-contextualizing an existing service (much like Twitter re-contextualized AIM’s away message functionality), or in intelligent gathering, aggregation and presentation of voices and users’ activity streams as they meaningfully relate to music.
