August 2, 2005 2:27 PM PDT
New Flickr tools rein in photo chaos
In the more than a year since its launch,
Flickr's early adoption of tags has made the service extremely popular because it is fairly easy to find images of, say, cats or Paris or the World Series by searching for photos tagged with those terms.
One of the new tools breaks down images with specific tags into clusters. For example, "turkey," which could refer either to the country or to the bird, is now divided into clusters for each possible interpretation.
The other new tool, called "interestingness," brings a Google-like relevancy to images by applying an algorithm that ranks photos--by date or by tag--according to how frequently the Flickr community has viewed them and how often they've been recommended.
"We wanted to build something that allows the good stuff to surface," said Stewart Butterfield, one of Flickr's co-founders. "We're relying on the implicit behavior and seeing what people are actually paying attention to, and doing the rankings that way."
While there are other photo-sharing services, few, if any, have built the kind of loyal following that Flickr has. That's due, largely, to the fact that its users have so many ways to make their photos accessible to friends and family, and also that its developers have concentrated on creating interactive ways for users to participate on the site.
But as the service has matured and as the sheer number of photographs it hosts has become more and more unwieldy, its developers have had to work on new ways to make the metadata (tags and comments) users attach to images more useful.
Flickr's power-users say the service has succeeded with its new tools, particularly the "interestingness" feature that allows users to see the photos others have thought highly of.
"It's great, because it works," said Matt Haughey, the founder of the popular technology culture blog, Metafilter, and a longtime Flickr observer. "The hardest problem to solve with any big service is to be able to answer the question, 'Can you show me just the good stuff?' Music, photos, you name it, that's a nearly impossible thing to automate and paying editors to find the best of doesn't scale."
Similarly, the clustering feature gives users a way to see, at a glance, photos organized in more meaningful ways than Flickr had previously offered.
"Prior to them perfecting the whole community metadata tag thing, this would be nearly impossible to do," Haughey said. "You'd have to do photo analysis to figure out if a photo was of the beach or some sky. That's like big supercomputer Xerox PARC type stuff."
Butterfield plays down the achievement, but he acknowledges that both the clustering and interestingness tools required a great deal of work.
"It's just a lot of really, really hard computation," he said. "For every tag in the system, we're looking for how often it coexists with another tag on the system. So that's just a brute force system."