October 12, 2005

Yahoo Aims To Be Research Powerhouse

Continued from page 1

By Wade Roush

Raghavan says the work of the research division is being organized into five focus areas:

Search and retrieval, including studies in what Raghavan calls "adversarial" information retrieval. "There is this ongoing, two-sided game between the Web search engines on one side and the spammers on the other," he explains. "You have to think two moves ahead of the spammer. You have to ask, 'If I make these fixes, here is how the fixes would manifest themselves, and the spammers are going to be able to infer this and that, so I better do this too.' It's an evolving game that's never going to end."

Data mining and machine learning. One illustrative project is Mindset, which is available for testing at next.yahoo.com, the company's public online area of research projects in progress. Mindset puts sliders alongside Yahoo search results that allow users to tailor search rankings to highlight either editorial or commercial content. Machine-learning algorithms use keywords and other clues inside Web pages to gauge their score on this axis. In principle, Mindset-like sliders could be used to adjust search results along any parameter.

User interfaces and user experiences, including social experiences using short-range wireless protocols such as Bluetooth for locating friends and potential contacts. The question that interests Yahoo in this area, according to Raghavan, is "How do you have a million or a billion people interacting with a million or a billion computers and other devices and create something better than a million individual experiences."

Utility computing. Raghavan's definition of the problem: "How do you spread a million PCs around the world and have them work as one big mainframe, and extend that beyond the PC into the [cell-phone] handset."

Microeconomics. Economic theory can illuminate many types of user behaviors and beliefs on the Web, says Raghavan. He says Yahoo would like to understand how to deploy incentives to reward honest users of social media such as My Web 2.0 and Flickr, sites where users can share Web sites and photos they've "tagged" with their own keywords. In October, Technology Review named Yahoo researcher David Pennock as one of the 35 most promising innovators under age 35 for his work on predictive markets such as Yahoo's Tech Buzz Game.

Although the research in all five areas is ultimately aimed at making Yahoo a more popular destination for Web users, not every project at Yahoo Research has a direct relationship to an existing product.

"There isn't in all cases a one-to-one mapping between our research areas and the business units that Yahoo has," says Raghavan. "But that's as it should be. We see ourselves not just as tactical problem solvers for this or that business but really as architects for the businesses in the making."

Microsoft's Rashid says he has two suggestions for Raghavan and Fayyad: "The most critical things you can do when you start a new organization are, one, hiring the right people, and, two, establishing the right values. If you don't exemplify the right values even the best people will fail. If you don't have the right people, no amount of organizational support will help."

Establishing the right values often takes time. Meanwhile, Yahoo's Raghavan is searching for the most valuable researchers.

"Developers involved in a software organization typically produce $1 million in revenue per year, per developer," Raghavan says. "If my scientists each produce $1 million in revenue, that's not very interesting. The thing I look for is for our scientists to think about hundred-million-dollar ideas or billion-dollar ideas."