An Interesting Idea for Flickr

There’s a good chance when I decide to upload an album of digital photos I’ve got many candidates of the same shot. I usually compare them back and forth until I can determine which one I think is best, but I’m not a professional photographer and I have little concept of how “interesting” any of the candidates might be.

But Flickr does.

Flickr is known for having its secret “Interestingness” algorithm which it uses to determine how interesting a photo is. It applies this algorithm to photos and if a certain threshold is exceeded for a shot, that photo is highlighted in Flickr’s Explore page.

Although their precise definition of an “interesting” photo is secret, we can glean a little bit of what Flickr thinks is interesting from their patent application for the feature (emphasis mine):

Media objects, such as images or soundtracks, may be ranked according to a new class of metrics known as “interestingness.” These rankings may be based at least in part on the quantity of user-entered metadata concerning the media object, the number of users who have assigned metadata to the media object, access patterns related to the media object, and/or a lapse of time related to the media object.

This leads me to believe the algorithm looks at both metadata features inherent in the photo like EXIF data, probably things like exposure, depth of field, etc., and social network metadata like how many times an image has been Favourited, added to a group, or commented on, and uses this in a recommendation system like Amazon’s “Collaborative Filtering”.

Even if that’s not precisely how Flickr does it, there’s a good chance such a mix would work well anyway. So with this algorithm, why don’t they help users decide which photos to upload?

Flickr’s algorithm would work really well here. I’d upload reduced versions of all candidate photos for my album, and Flickr would analyze them, returning to me the set it’s determined are the most interesting. If Flickr can’t determine which photos are candidates for the same shot (although it should be able to determine this based on contents and time stamps) then it could let me group them together and it can pick the best of that group.

This way, I can then upload the photos Flickr thinks are the best and I put my best foot forward for my photos. This isn’t about me getting more “Favs” or comments on my photos, this is just about making use of an excellent algorithm to help me, a non-expert, determine which of my photos are probably the best, because I’m not so great at doing it myself. Over time, this might help reveal patterns to me on how to take better shots (Flickr could also help educate users if that was their concern).

If Flickr has an algorithm to do this, why do I still have to do it manually?

I felt this applied best to Flickr because they’re in the mid-range of Photography websites. Facebook is a dumping ground for all your photos and 500 Pixels is generally used by professional photographers who already have a better sense of which photos work best, but they might stand to benefit from this method, too. Ultimately, I’m a Flickr user and they’ve already got the algorithm.

Speed of Light