Feb 1 2008

Google Image Labeler, Human-powered search

Posted by cheyennejack at 7:48 PM
0 comments
- Categories: SEO / Marketing

Share this: del.icio.us digg StumbleUpon Facebook Technorati Fav Mixx Spurl newsvine Furl reddit Yahoo! My Web Wists Simpy BlinkList BlogMarks smarking Ma.gnolia MSN Live Google Bookmarks Segnalo diigo excites Netvouz

I played around with Googles new attempt to add human-powered searching results to their images at the Google Image Labler page.

I can understand the idea. Kind of reminds me Jason Calacanis and Mahalo's Human-powered search idea.

While its interesting that they turn it into a game giving you and some random partner 2 minutes to guess at which keyword is most likely, I do think there can be some flaws to this methodology.

1) Because it is a game you become competitive in trying to beat the game rather than solve the problem. For instance, if I see a picture I know is Quarterback Tom Brady, then thats what I should type. However, after doing this for a time you find your counterpart might type "guy, man" or worse yet the color of the shirt he is wearing. This teaches you bad habits down the road and soon pictures of Tom Brady get associated with the color "red".

2) Along the same lines, you start to realize that singular words are more likely to match your teammates guesses who thinks the same thing and does one word as well.

3) They do start to eliminate certain words by marking them as not valid and showing them to you, similar to the game Taboo. Problem being that with Taboo you can't say that word or any deritive. So you learn in the Google game if you can't say "football" or "player" you type "football player" and suddenly get a match. I even found myself reading the words I couldn't say first before looking at the picture.

4) Quality. The quality of the images are very poor. Sometimes they just show you random cartoons you can barely make out. They need to index only higher quality images that you can tell what it is anyhow.

5) Text. Finally if there is any text written in the picture, you learn to type that first whether its relevant or not. If someone is in the middle of a fight wearing a shirt that says "peace". You type "peace" and you get a match since your counterpart figures out the same thing.

Granted I'm sure this won't be the only data they use in building the image search, but still something to consider as human-powered search becomes more popular. Games are good idea to entice, but it becomes a problem when beating the game is a different exercise from the end goal of producing relevant searches.

Share this: del.icio.us digg StumbleUpon Facebook Technorati Fav Mixx Spurl newsvine Furl reddit Yahoo! My Web Wists Simpy BlinkList BlogMarks smarking Ma.gnolia MSN Live Google Bookmarks Segnalo diigo excites Netvouz




Comments

Write your comment



(it will not be displayed)