Google’s Image Labeler Helps To Refine Image Search
One of my favorite scenes in Ghostbuster’s, is where Bill Murray is interviewing a really hot undergrad for an ESP test and even though she gets all of the answers wrong she still gets rewarded. Meanwhile some poor guy gets electrocuted every time he answers. That’s almost how I feel about Google’s new image labeler game. Basically, Google matches you up with a random person and then shows you images that are contained within their search results. Both players start tagging the photo as fast as they can until you both end up matching a tag. Once my tag matches my partners tag, then you go on to the next photo. It’s a lot of fun, but frustrating and humbling to say the least. Search Engine Watch has a write up on the service and asks if this game will really be a threat to Flickr. I think that this does start to poach a bit on Flickr’s turf, but at the same time it’s really only an image refining tool and doesn’t offer the social aspects that Flickr does.
The game itself wasn’t really made as much for players to goof off with, but is actually a research tool for Google. I’m not sure how exactly they are filtering the results, but the idea is that if people always identify certain photos with certain things then it will improve the ranking on that image. I’ve played the game a few times and have had various results with it. My first partner only seemed to want to make one tag while I started tagging as fast as I could. I figure that if I list ten items it’s a lot easier to find a match if my partner lists ten items as well, but this guy must have thought that he only got one guess and I just couldn’t come up with it in the 90 second time limit.
The second time I played I must have gotten someone who thinks very differently then me because no matter what I put down, it was only the more obscure terms that would match. For example, there was a photo of a group of guys standing around smoking. I tagged people, guys and smoking, dudes, sunglasses, gang, group, etc. but the tag that matched up was wife beater because one of the guys had a tank top on. The entire game itself really makes me think about how I use language and how search is restricted by the searchers own language barriers. Words can be tricky things and even the most well read individuals can still have trouble coming up with the most accurate description of something. In the case of Google’s image labeler, I think that it will help results, but I don’t know that accuracy will be the final result. Instead I think that it will have a bias so that the most popular tags gets associated with the right photos and not necessarily the most accurate ones. When I saw a photo of the Aurora Borealis come up, I tagged it such, but it was really northern lights that my partner and I agreed upon. On one hand because we agreed, maybe northern lights is a more refined search, but Aurora Borealis seems like a more accurate description to me. Nontheless, the game will definetely improve Google’s image search by having a human look over the photo and recognize things that their search tools just can’t. In the end it’s a great way to kill time and it’s helping to improve the Google search results.
Posted on September 3rd, 2006 by Davis
Filed under: Search