In this post I’m presenting a site that I developed in my free time. It served me to put my PHP skills into work, and to kill some free time while I was working abroad (free time which, sadly, I don’t have anymore).
I had the idea of building ourlooks.com after watching this thought-provoking talk by Luis von Ahn, where he explains some of his research activities. The basic idea behind the talk is: we can automate tasks which computers can’t nowadays solve by making humans work for us. And humans will work, and they will do it for free, as long as they have an incentive to do so. More interestingly, this incentive is having fun. So, von Ahn has built several games that people seem to enjoy - and while playing, they solve interesting problems, such as recognizing the objects inside an image or collecting general knowledge (e.g., bananas are yellow and cars have four wheels).
I thought about other ways this idea could be applied, and this is how ourlooks.com was conceived. One thing that computers can’t do is to tell you how you look, especially in terms of how other people perceive your personal image. So I decided to create a game that, while being played, served to rank people in categories such as looking pretty, trustworthy, and others.This initial idea is what later on derived into what ourlooks.com is today.
Sadly, ranking user’s looks was not a 100% new idea. Most notably, the people at hotornot.com have been making millions of dollars since they were the first site (at least the first that got to be widely known) where users could vote on other people’s personal image. There exist nowadays tons of clones, devoted to rating cats, nude bodies, similarity to celebrities, and many others. However I thought that ourlooks.com would be an interesting experiment as it was significantly different: instead of hosting a popularity contest, it could become a useful tool to assess your personal image (you can have your own opinion on whether this is true or not).
Picture ratings
In ourlooks.com, every user-submitted picture is automatically rated in the following categories: pretty, trustworthy, happy, intelligent, funny and hot. All kinds of combinations are theoretically possible: someone can be hot but not pretty, or funny and not look intelligent.
The rating mechanism works in the following way. When a visitor enters the site, a question is displayed, followed by two pictures belonging to two different users. There exists one question for each of the previous categories. The question serves to tell the user what is the category under evaluation: “Who looks more intelligent?”, “Who is the hottest?”. Based on the votes from people, every picture gains or loses points in the corresponding category.
Most sites ask the viewers for a direct numeric rating of each picture. This adds more variability as the marking standards of viewers are different. By asking the viewer to vote over two pictures, ourlooks.com just captures the preferences of the voters and generates the scores indirectly. Also, comparing two pictures should more convenient for the user: selecting a numerical value requires more mouse movement than just clicking one of two pictures.
ourlooks.com uses a mechanism that resembles that of the ELO rating system used to rate chess players. When a low-rate picture (weak) is prefered by the voter in front of a highly rated one (strong), it will gain many points. Conversely, a strong photo will not gain much for an easy win in front of a weak opponent. This makes every picture’s rate to be quickly adjusted after a few votes. Allegedly, since I have not formally done any research on this.
Every picture that the user uploads is rated independently. For example, the following pictures are of the same user. Each photograph is obviously to be rated very differently.
Picture search engine
Besides the obvious ability to show top performers in each category, the site includes a unique feature: the search engine, which can be used to locate pictures depending on the ratings in each category. To my best knowledge, this is the first site user search engine that can take into account the looks of the people.
The user can specify the characteristics that he or she expects from results. For example, the first result of a query for “very pretty but rather unhappy” people returns these pictures in the first results:
And “happy, trustworthy” returns these results:
MySpace widget
The last feature that I added to the site is the “myspace widget”. Every user has the chance to include a picture in their myspace (or similar) profile that points to ourlooks.com. The picture is dynamically updated with the current ratings. However this is something that the users generally have not put into use. I guess they’d rather keep the results private.
Conclusions
The ourlooks.com site has now gathered around 500 registered accounts, and users have uploaded around 1000 pictures. Interestingly, most of the users are women. I guess women have a higher tendency to be worried about their looks. Either this, or the text ads I used to promote the site were unwittingly targeted to women.
For now the site is a small club, so I am happy to receive feedback from users. Too bad that I can’t dedicate much time to make it grow in features.
Post a Comment