Since Flickr is one of the most well-known Web 2.0 sites it is worth taking a look at what they do if for no other reason than to use it as a predictor for what the legions of Flickr clones will soon try to copy. Today we are looking at Flickr's method for selecting what it terms "interesting" photographs–purportedly without the intervention of human editors. The results are generally pretty impressive adding to the question what variable they use to distinguish compelling visuals.
"More than just "interesting," "interestingness" could potentially be a way that Yahoo! reclaims a little piece of search from Google. Today image search at both Google and Yahoo! is largely broken. Do a search for "San Francisco" at both Google and Yahoo! Image Search and you will find a hodge podge of mostly mediocre images.
"What’s great about this is that it exposes not the wisdom of the crowd but the taste of the crowd"
Now for the algorithm. Interestingness is described on Flickr as:
"…an amazing new Flickr Feature.
There are lots of things that make a photo 'interesting' (or not) in the Flickr. Where the clickthroughs are coming from; who comments on it and when; who marks it as a favorite; its tags and many more things which are constantly changing. Interestingness changes over time, as more and more fantastic photos and stories are added to Flickr."
Emphasis has been added.
By hinting at the existence of secret sauce, Flickr enters the "we're more than a pretty face" and "trust us we have amazing algorithms underneath" competitions along with people like Digg. If fact it's getting to the point that if you can't work the word algorithm into your "about us" you risk being called "web 1.0"!
Looking at Flickr's description of interestingness above you get a hint at what gets a photograph selected for this distinction and here is what we (and others) have been able to determine.
- Views, internal and external to Flickr, of the photo
- Number of comments on the photo, and also who comments on the photo
- Tags applied to the photo
- Flickr discussion groups in which the photo appears
- Favorites, a.k.a Flickr bookmarking, of the photo
- Time varying behavior of the above factors
Flickr mentions "who" twice in their one paragraph description of their process, but that is a much more difficult aspect to deconstruct. However one blogger noted an absence of the amaturish photos that seem so prevalent and wondered out loud how this might be so:
"One conclusion to draw might be that the professional and semi-professional photographers who make up a minority of Flickr's users are having a disproportionate influence on the metrics that go into Interestingness because they are more active. They make more comments, mark more photos as favorites, look at more pictures not by their current contacts, and therefore their activity has a greater weight in the algorithms that choose the Interesting photos."
Flickr appears to have tinkered with their algorithm (in Feb?) and introducing a penalty for those who appear to try and game the system by uploading to numerous Flickr groups. Aocording to one Flickr user:
"…some recent changes to the algorithm devalued the interestingness of photos submitted to too many groups. This had sparked controversy with a specific kind of Flickr user affectionately referred to as a 'group whore'."…"Group whores are users who send their photos to tons of different groups in a desperate attempt to garner attention (read: views, favs and comments) which in turn would hypothetically lead to a higher level of the coveted interestingness."
Here is a debate between flummoxed Flickr users and the Flickr founder over the issue.
Flickr is right, it is an amazing new feature.
Update: Thomas Hawk adds this hypothesis to Flickr's Interesting algorithm:
One major change that has also occured with regards to interestingness (in my guestimation of course) is that averaging has been introduced for more popular photographers to prevent them from overly dominating interestingness.