Edinburgh Printmakers, Edinburgh, Scotland
Originally uploaded by beauvaislyons

Part of my on going vanity searches I found that this “Photo from Alastair Clark” appears third on a Flickr search sorted by “Interestingness”.

I gather the algorithm, as I’ve mentioned on my blog before, must value things such as Photo Views, Favourites and Comments. This photo has been favourited once, got one comment on it and been viewed 48 times as of 13th December 2009. However two of my own photos performs better and these have neither been favourited nor got any comments.

The only differences are the tags, person labels and photo views with one of the two photos having 51 views, the other 39. Maybe the algorithm includes weighted tags, or number of tags, but it’s hard to see how that can overcome the human element of higher views, favourites and comments. I doubt that the interestingness algorithm does any analysis of the photos to value such things as composition, colour spectrum and content as this would probably require too much calculation.

I’d be impressed if Flickr could handle that, given the fact it receives huge numbers of visitors daily, along with handling all the uploads etc. The actual data storage requirements wouldn’t be all that massive, and it’s not like the stats data they used to collect about search terms and linking sites, which would be forever growing. After the calculation was done once the set of values about some key attributes would never grow in size, unless the algorithm was changed and new attributes were added.

This posses another problem, with all the pre-existing photos. How do you go about calculating and recalculating an interestingness value based on the actual image when you have such a huge database of photos. It makes sense to be able to process new uploads, and perform calculation on them, that’s a matter of through put. But re-evaluating all the existing photos whenever the algorithm changes would just take forever and doesn’t seem very practical.

Thankfully the picture doesn’t do so well on relevance, as tags are probably weighed higher than photo descriptions in which, for this photo, my name appears.