O’Reilly Netowrk‘s Richard Korman has a fascinating interview with Flickr’s CEO Stewart Butterfield. (Love that name, by the way.) There’s lots of good stuff here about what makes Flickr work as social software and how people are using it. Here’s what Butterfield has to say, for example, about folksonomies:
Koman: About the tagging, Flickr and del.icio.us are usually mentioned in the same breath of having pioneered this concept of “folksonomies.” Adam Mathes calls this a shift in who creates the metadata from the author to the user. Lots of del.icio.us and Flickr users describe an object–a web page or a photo–in the way that makes sense to them, with no interest in how the author might have classified the page.
Information experts like Lou Rosenfeld complain that folksonomy tags, while interesting, don’t support searching or browsing particularly well, don’t support relationships, don’t offer true classification, and can’t even cluster synonyms. So why does this uncontrolled mass-tagging by users makes sense?
Butterfield: First, going back to the del.icio.us comparison for a second, we were definitely directly inspired by del.icio.us. There’s an interesting difference, though: because people are adding URLs to del.icio.us, and many people can add the same URL, you end up with multiple ways of tagging the same thing–different people’s vocabulary for the same item. On the other hand, each photo a user uploads to Flickr is unique and belongs to them; however, more than one person can tag it. (For people who have large personal networks in Flickr, they definitely find that they upload a photo and go to sleep, and in the morning there are tags all over it.)
The complaint that it’s uncontrolled and it’s not going to be captured in a consistent way to me is really irrelevant. Because tags are first and foremost for people to organize their own photos–and if they weren’t, it wouldn’t work. It’s a happy accident that the whole global collection emerges. And let’s say it’s only 50 percent accurate and complete and let’s say right now we have 10,000 photos tagged “Italy;” it might actually be 20,000 photos that should have been tagged “Italy,” but who cares? No one is going to look at all 10,000 photos, let alone 20,000 photos. And in six months, it will be 50,000 photos instead of 100,000 photos.
Now there are some things that will annoy people, like singular versus plural forms, homonyms and homographs, and alternative spellings. But a lot of this is easy to deal with when you have enough data. In general, disambiguation is a harder problem.
We’ve done step one of relatedness of tags, which is just cluster analysis of how people tag, and then we suggest related tags. So if you tag “Italy,” it will suggest “Rome,” “architecture,” “travel,” “food,” “Europe,” etc. And it works astoundingly well. People usually think there’s a human editor, but it’s just cluster analysis. And there is a lot more of that coming–some of which we’re hoping to show at the 2005 [ETech] conference.
Yummy!
