Wednesday, August 20, 2008

Pick the transformer

I finally got around to making a fun little website where you are shown two classic Transformer toys from the 80's and you have to pick which one you get to keep and which one ends up getting traded away. The site is called called Pick the transformer. It keeps track of the votes and is able to rank all of the toys and show some basic stats on the votes. It is fun to play with and easy to find yourself clicking and clicking.

Back in 2000 a site called amihotornot.com (now hotornot.com) was launched. The site showed the viewer a photo of a girl and you got to rank her 1-10. While not a perfect system it was simple enough to get its internet moment of fame. A little while later another site called pickthehottie.com was put up that showed two images and you had to select which girl you found more attractive. Girls submitted their own image because they want to see where they rank and guys, well they want to look at girls. The sites have expanded sense then to have all sorts of options and extras, but at the core it is just trying to rank photos. The algorithms between the two sites are very different and believe that pickthehottie is much better then hotornot.com. On hotornot you are faced to rank a photo between 1-10 with no guidelines as to how to make judgments. While it is simple enough to say that photo X is near the top if most votes put it at a 10, you can't know with very good accuacy where the photo ranks in relation to the other photos that are mostly 10. (or another way who is the hottest?) This is further problematic when you consider that most photos will cluster around several numbers such as 1, 5, 7, and 10. To rank anything from 1-10 you need to first get a metal baseline of what are the two extremes of 1 and 10. The first few images you have to vote on are very speculative and it can't be until you have seen a number of images that you can consistently rank images. But even once you are consistent you must constantly try to compare the current image with all previous images you have seen which can cause a 7 to become a 9 or 5 over time as you get more and more input or just change your opinion. Contrasting this with pickthehottie I think that many of those problems go away. Given the choice between two images users only have to pick the one that they like more. This is usually so easy that users don't even think about why they think one photo is hotter, they just know it. It is only when faced with the choice between two very similar photos that the user would even have to make a hard choice. Behind the scene every photo contains two lists of photos that are hotter or cooler then the current image. Just based upon the percentage of the two votes you could generate a ranking for each photo. Taking it one step further you can fine tweak the ranking by only counting the votes that were places when comparing two photos of similar percentages. This was a fun little two day project that came together very quickly and is fun to use.

No comments:

Popular Posts