« Matt Locke on folksonomies |
Main
| Vimeo - tagged video »
March 1, 2005
Popularity Slider: Diving into the long tail
Posted by Seb Paquet
The general idea of a recommender system is that it asks for a few examples of things you like and then gives you more things it thinks you might like, based on its knowledge of other people’s preferences.
One problem you can often run into when using a recommender system is a bias towards popular items, which are not really that close to what you like but have the favor of many users because of their high visibility. For instance, based on my subscriptions, the Bloglines recommender keeps suggesting that I have a look at Slashdot, always putting it near the top of its list of suggestions. The effect of designs like this, of course, is is to reinforce the “short head” (as opposed to the “long tail”) by directing users towards the roads well traveled.
An easy way to mitigate this is to selectively decapitate the recommendation engine’s results. Last year I blogged about Andrew Grumet’s “Similar Feeds”, which implements this. I just came across a music filtering site that makes the feature more prominent and intuitive by putting a nice, fat “popularity slider” right at the top of recommendations pages. Try playing with the slider on this page to see how it works.
I like how things like this underscore the idea that “this is popular” is not the same as “you’ll like it”.
Comments (7)
+ TrackBacks (0) | Category: social software
- RELATED ENTRIES
- › Spolsky on Blog Comments: Scale matters
- › "The internet's output is data, but its product is freedom"
- › Andrew Keen: Rescuing 'Luddite' from the Luddites
- › knowledge access as a public good
- › viewing American class divisions through Facebook and MySpace
- › Gorman, redux: The Siren Song of the Internet
- › Mis-understanding Fred Wilson's 'Age and Entrepreneurship' argument
- › The Future Belongs to Those Who Take The Present For Granted: A return to Fred Wilson's "age question"
1. Paul Madsen on March 1, 2005 1:26 PM writes...
Hi Seb, I'd like to see this sort of slider personalized. For instance, if I'm on a music site, I should be able to indicate that today I want to listen to music 'different' from that which my previous selections/playlists would indicate - making it easy for me to occasionally splash around in the shallows of my tail. This sort of exposure to new genres/artists is something that my real-world friends have always provided me
Permalink to Comment2. Peter Hoskins on March 1, 2005 3:51 PM writes...
Here we have a self perpetuating problem. As more and more sites/blogs/content get created, it gets harder and harder to sift through the noise to find the nuggets. So...it gets easier and easier to just go to the slashdot type sites and not waste time filtering through the long tail. And so the log tail just keeps getting longer.
Recommender systems are quantitative not qualitative. The popular blogs are recommended because they are popular and that means they are good - right? Slashdot is qualitative because there is some editorial judgement being used. Until recommender systems add some kind of qualitative attribution they will only reinforce the status quo. But if they did add a qualitative attribution then wouldn't they have come full circle and become exactly what slashdot, corante, scripting are today...?
Permalink to Comment3. Julian Bond on March 1, 2005 4:16 PM writes...
Compare Amazon's music recommendations here with last.fm's take on this. Amazon is seriously lacking in serendipity, where the more random last.fm constantly throws up new and interesting results. The end result is that Amazon gets narrower and narrower as you tell it more about yourself. While last.fm gets a better understanding of people like you but gets to choose random results from a larger base.
Permalink to Comment4. Shannon Clark on March 1, 2005 6:02 PM writes...
I've never had a single recommendation on Bloglines (probably some bug - I subscribe to 150+ feeds there, should be enough to get some real, useful recommendations?)
In general though I agree that there is a need for context and variation in recommendations - the classic problem being "buying gifts on Amazon" - once you do, you no longer get recommendations specific for you. However, you might be very interested in recommendations that took into account your friend's wishlist and your past purchases for them (possibly even correlated with your own past purchases/reviews).
Permalink to Comment5. Toby on March 1, 2005 6:40 PM writes...
For the record, Mobster had the popularity slider long before upto11 did.
Permalink to Comment6. Richard Akerman on March 6, 2005 9:11 PM writes...
Hey, it happens I have just been looking at music recommendation engines. Here's what I found
http://blog.akerman.ca/2005_03_06_archive.html#111015173334585906
Permalink to Comment7. dan hartung on March 9, 2005 6:06 PM writes...
I definitely agree with the Amazon problem; I blogged about that a couple years ago, in fact. I was trying to use their reccos as an exploration engine, but it was geared toward sales rather than my needs. Sure, it makes sense that if I own Beck's Odelay I would be a good bet to buy Sea Change or Guero ... but maybe that's not what I'm looking for today.
The popularity slider is a really good tool to have available. Another implementation is Yahoo! LAUNCH, which lets you browse bands via "More Popular" and "Less Popular" controls. I actually like that one a little better, because (gasp!) sometimes I like to cut off the long tail, because all those results can be overwhelming and noisy.
Permalink to Comment