Perils of web taste personalization
Melody
Nice variations, but nothing particularly disruptiveI am exhausted of Amazon's recommendations to me. And I'm tired of hearing the same songs continuously on Pandora and seeing such similar music recommended to me by iTunes Genius. In short, taste personalization is feeling constrictive and extremely limited.
It wasn't always this way. Not two months ago, I was writing about how much more music and books I was buying thanks to the wisdom of smart algorithms of recommendation sites. Now I'm beginning to realize that only so many bands sound like The National and there are really only so many books like Nudge that I want to read. In the beginning, "if you like this, you'll also like this" can be a powerful way to find new artists and authors, but eventually this approach limits rather than broadens my perspective.
The value of independent curators of content is rising in my mind. The New York Times 100 Notable Books of 2009 led me to try out Death in Venice, Jeff in Varanasi, a piece of fiction unlike anything in my Amazon browsing history, and I loved it. I'm also now working my way through much of the Modern Library's 100 Best to find recognized, time-tested classics rather than just stuff that I like. I trust the collective (and attributed) criticism of Rotten Tomatoes and ignore Netflix's recommendations. For music, Pitchfork is not a bad option, although my most trusted curators are usually drawn from friends. My father, who works with last.fm, recently argued that I should abandon Pandora if favor of that site because social networking approaches to discovery are more powerful than algorithms. I still would probably just prefer to ask what he is listening to these days.
Smart recommendation algorithms have their place, but I look to music and books to expand my outlook, and I'm realizing that the human connection, whether from friends or prestigious curators I trust, is a better way to get there.
Flickr Creative Commons credit: Laurence and Annie
Reader Comments (4)
I wonder if this points to deficiencies in The Long Tail. Anderson wrote his thesis before iLike took off, before our friends announced their credit card purchases through Blippy. I think personalization services as currently architected reinforce homogeneity of interests. They support new discovery, sure, but with 99% of the signal clustered at the hits. As you allude, the gravitational pull of The National has its own echo chamber just around its periphery.
But maybe there's room for Pandora and its cousins to mature. The binary simplicity of "thumbs up" and "thumbs down" is seductive, but I think we can handle some complication: "I'm feeling adventurous"; "I hate all of my music right now"; "I was barely listening anyways".
P.S. Recently I've liked Okkervil River's The Stage Names. "The Song Is You" by Arthur Phillips was a fresh read.
Look what else they're trying to do with algorithms. If you enjoyed your date with Tad... http://www.nytimes.com/2010/02/07/business/07stream.html?src=twt&twt=nytimesbusiness
@Joseph: I love your examples of more complex feedback and I, too, fear that these sites reinforce taste homogeneity. And thanks for the recommendations! I've never heard of either and in this case, that is just perfect :)
@Kurt: ScientificMatch.com and Chemistry.com are disturbing takes on recommendation algorithm worship. Thanks for the pointer.
The better algorithm is called the Bayesian Truth Serum. read my blog post and ask your Hunch.com friend about it. i have it modeled out in excel if you want to see the formula. just leave a comment on my post and i'll email it to you (in case i don't come back to this specific page)