Online games aid image search

BANGALORE: Whoever said online games are a waste of time? Ask search companies Google and Yahoo! and they are surely bound to disagree.

With more and more user-generated content and rich media such as photos, music and videos flooding the Internet, traditional search technologies are not able to pull out the ideal mix of search results.

In such a scenario, search companies are now relying on online collaborative games to study user behavior that could give them insights into enabling better image and video search.

Andrew Tomkins, VP, search research, Yahoo! Research who specializes in social media search is of the opinion that traditional search technologies fall very short when it comes to looking for rich media content.

“Traditional search allows us to get the URLs along with abstract. However, we haven’t reached a stage where we can pull out podcasts, images or blogs along with the URLs, when one searches. We are still in the early days of that. Ranking of multimedia objects is very difficult. In video and image search, the relevance of the object is around pixels, which is hard,” he said.

Traditional search technology crawled the HTML pages and pulled out the relevant pages. But in the social media world, there is fragmentation of content types. To overcome this, researchers are banking on data generated from games that give an insight into how users provide cues for search.

One such popular game is the Extra-Sensory Perception (ESP) game developed by professor Luis von Ahn, an assistant professor at the Carnagie Mellon University, USA. In this, a player is pitted against an unknown player and both have to provide labels to a random picture that is provided. A player scores points when the labels for the same picture are common. The end goal is to type as many labels possible within the time limit till the two players find a match.

Von Ahn spells out the reasons why image search is so challenging. “It works by using filenames and text on websites. So, for instance, when you search for the query "dog", you get a lot of images that have the word "dog" as their filename, or very near to them on the website. This doesn't always work very well since there is no guarantee that the filename is related to the image. The games can help in providing ground truth for a large number of the images. Around 250,000 people have played the ESP game so far,” he said.

Tomkins who is excited by this new trend in search, feels that the game is really about a convincing a massive collection of online users to tag images. Google, which was impressed by Von Ahn’s game, has licensed it and offers under the name Google Image Labeler. Von Ahn has also developed other games such as Peekaboom and Verbosity.

He is now working on a new set of games to solve other problems such as music labeling, and language translation.

©CyberMedia News