Statistical Pop Art II

This is a follow-on to this post.

As I suggested in the previous post, you can do some of this stuff with metadata, like tags. A very simple approach is to look for unusual tags, simply pick random words from a dictionary after eliminating common tags. A slight refinement would be to search for photographs with at least one tag, but which is not tagged with any of, say, the most common 1000 tags.

This doesn't reliably filter away common photographic themes and ideas, and it doesn't reliably filter in awesome and interesting new ideas. That doesn't matter, we're taking a purely probabilistic approach here, anyways. All we can really hope to achieve is an increased density of work that's interesting in some way, and uncommon in some way.

I have actually performed the experiment of looking for randomly selected unusual words on flickr. It definitely produced search results that were less dreary than, say, flickr's Explored Images, or whatever. I don't think I really uncovered any interesting communities of artists creating basically new work, but I did uncover some individuals doing things I liked.