Константин Сонин (ksonin) wrote,
Константин Сонин

Из серии "Великие соавторы"

В Wall Street Journal вчера вышла статья про моего соавтора Майкла (Михаила, конечно) Шварца. С его портретом на первой странице! Ничего себе, да? Побыв профессором в Гарварде (там мы написали нашу статью про аукционы), проведя год в институте Гувера в Стэнфорде (там мы закончили нашу статью, которая сначала называлась "Война и мир" - она теперь выходит в JLEO) и два года в Беркли, он переходит в Yahoo!, чтобы стать там первым профессиональным экономистом. Он становится одним из самых известных специалистов по интернет-аукционам в мире.

WSJ, August 25, 2006; Page A1
Lab Test
Hoping to Overtake Its Rivals,
Yahoo Stocks Up on Academics

Economists and Search Gurus
Fill New Research Team;
Data-Rich Fantasy Land
Looming Privacy Concerns

Michael Schwarz earned an economics doctorate from Stanford University before spending five years as an assistant professor of microeconomics at Harvard University. His research papers include, "Synchronization Under Uncertainty."

These days, Mr. Schwarz is thinking about how to use economics to save attractive women from unwanted solicitations on an Internet dating site. One idea employs the concept of "scarcity," rationing the number of free messages each lothario can send. Another uses full disclosure, by displaying how many people a suitor has already approached.

The Russian-born theorist, 36 years old, is at the leading edge of a push by Internet giant Yahoo Inc. to beef up its business by hiring prominent academic economists and other researchers. The company, which offers consumer services ranging from email to fantasy sports, is betting that sponsoring fundamental research will help it tackle some of its biggest challenges, including battling the technological prowess of Google Inc.

Yahoo was blindsided by Google's sophisticated Web search. One of Yahoo's advertising-sales techniques also underperforms its rival's, and when Yahoo said last month that a revamp would be delayed, the company's stock fell 22%, its largest-ever one-day drop. Despite having one of the world's biggest user bases, Yahoo hasn't fully benefited from hot phenomena such as online video and social networking, a service offered by sites such as MySpace.com.

The research push, "has huge consequences for the business if we do things right," says Usama Fayyad, Yahoo's chief data officer.

Central to Yahoo's goal is its ability to record what millions of consumers do every day, and to study how changes to the company's Web services affect their behavior. Internet companies in the past have largely lacked the systems and focus to mine data for research, but now they're viewing it as a key competitive pursuit. For economists, Web operations are data-rich fantasy lands where they can observe in real-time the behavior of millions of consumers in varied marketplaces far more effectively than ever before.

One potential obstacle to collecting and analyzing a vast amount of data is customer privacy, particularly in the wake of concerns stirred up by Time Warner Inc.'s AOL unit earlier this summer. It inadvertently released a slew of information relating to users' search queries.

In addition, tech companies have a mixed record of translating research into profit. Xerox Corp.'s Palo Alto Research Center is widely credited with inventing several key features of modern computing in the 1970s. But it was Apple Computer Inc. and others that capitalized. Google, with such cautionary tales in mind, sprinkles researchers through its product groups, supplementing a small, standalone research unit. Some Yahoo staffers question whether the company's engineers have the time or inclination to implement ideas from the research team.

A Yahoo spokeswoman won't say how many economists or researchers the company plans to hire. A person familiar with the matter says Yahoo aims to build a team of more than a dozen economists. In the past year, the company has snagged leading talents in microeconomics, Web search and artificial intelligence from universities such as Cornell and Carnegie Mellon, and industrial labs including those of Microsoft Corp. and International Business Machines Corp. It has opened lab facilities in Berkeley, Calif., New York, Barcelona, and Santiago, Chile, and has begun scouting locations for a lab in Asia.

Until now, university economists typically have had to rely on limited sets of historical data to test their theories. At Yahoo, they can tweak market conditions and watch what happens to the company's roughly 500 million monthly visitors. The result is faster, and probably more useful than traditional methods such as surveys. Researchers inside and outside these companies also believe collating Web activity can be used to predict future events, such as box-office openings and home sales.

"It's an amazing opportunity to validate theories that have been floating around for a couple of decades," says Cornell University Computer Science Professor Jon Kleinberg, who cites as an example ideas about how innovations spread throughout society. Prof. Kleinberg has collaborated with Yahoo's researchers.

The big push into research dates to March 2005, when Yahoo Chief Executive Terry Semel confided a concern to top lieutenants that the company risked missing long-term developments. Who in the company is considering what might happen in three to five years, he asked, according to one executive at the meeting.

Mr. Fayyad, 43, the company's head of data, is leading the effort to answer that question. In April 2005, he sold Mr. Semel on an ambitious plan to fund fundamental research. Mr. Fayyad recruited Prabhakar Raghavan, a former IBM Research veteran who was also wooed by Google and Microsoft, to head the push. Mr. Fayyad told Mr. Raghavan he would be disappointed if Yahoo's effort didn't produce a Nobel Prize.

Messrs. Fayyad and Raghavan agreed Yahoo Research would tackle just a handful of areas, including search, the interaction of users online and an artificial-intelligence field known as machine learning, which involves analyzing past data to learn about the future. "The intent isn't to gaze at the skies," says Mr. Raghavan, 45.

Instead, researchers are looking at the information Yahoo collects about its users' activities. That starts with nearly every click and includes what services consumers use in what sequence and which color of buttons they click on most. The company records over 12 terabytes of data daily -- the equivalent of about half the information contained in all of the books in the Library of Congress, according to some estimates.

Yahoo and other Internet companies already use some of what they know about consumers' online habits to target Internet advertising: A user who searches for "Ford Explorer" might see an ad for a sports-utility vehicle when he looks at a news Web site.

Mr. Fayyad co-founded a startup in 2000 that became Revenue Science Inc., a firm that specializes in such behavioral-targeting technology. The technology observes what a consumer does on a site in order to decide which ads to display. One trick is to generate good matches -- looking at an article about an ax murderer doesn't suggest the reader is interested in buying axes. Another challenge is not to creep out users with blatant matches that serve as a reminder they're being tracked, such as switching all the ads to those for credit cards if a user clicks on a card offer.

As the recent AOL firestorm highlighted, there are big privacy issues in this line of work. AOL apologized when its data, which were originally intended for use by outside researchers, were made available on a public Web site. While individuals searchers weren't identified, the words they searched for -- such as their names -- allowed some to be tracked down without much effort.

Yahoo says it has multiple protections to guard against an AOL-type mistake. The company says it strips out any personal identifying attributes before handing the information over to researchers. In any case, it says, Yahoo's researchers are interested in the behavior of mass groups, not that of specific individuals.

Guarding against violations is a group that's known internally as the "Paranoids." T-shirts worn by Paranoids at one point read: "We worry so our users won't have to."

Yahoo submits all outside requests to look at the data to a review committee. It says it has never provided search-related data to outside researchers. Information it has shared include data related to music preferences, based on millions of anonymous ratings.

Technology companies have rarely before placed such a focus on fundamental economic research. Microsoft Research, which employs 700 researchers, says it has no economists on staff. Lucent Technologies Inc.'s Bell Labs has a team of about 30 economists, but their work mostly involves using theory to solve nuts-and-bolts industry problems.

"Right now we're at the golden age where there's a lot of work that's going on that is really lending insight into real-world problems," says Hal Varian, a University of California, Berkeley, economics professor, who is a consultant to Google.

In 2001, Yahoo started charging for its online dating service -- which surprisingly resulted in an increase in membership. Mr. Raghavan thinks the switch increased the value of Yahoo Personals in the eyes of consumers and encouraged them to use it more. While the move predated Yahoo's recent research push, it's an outcome economists might have predicted: The fee deterred users who weren't serious about dating, making the service more efficient for those who were.

Mr. Fayyad says Yahoo Research has already more than covered its expense with a project to better target Yahoo's advertising for its own services, such as Yahoo Mail and Personals, on the Yahoo site. The team tweaked the algorithms that decide when to show such ads, which beefed up the response rate.

Yahoo believes research might help it with other Internet services whose design, and success, are currently determined by intangible factors. With the help of the research team, Yahoo designed an incentive system for Yahoo Answers, a service where consumers post questions, to which other users respond. Under that system, users receive points for answering questions, and get more points for answering questions that are well received. The top-ranked answerers are featured on a leaderboard and the ensuing competition has improved the service's quality.

Yahoo is continually tweaking the rules for points, such as by raising the penalty for people asking profane or abusive questions.

One of the team's biggest challenges is finding ways to improve Yahoo's ad-auction system, in which advertisers bid to have their ads shown to users who search for relevant keywords, such as "Orlando hotel." The advertisers pay each time consumers click on the ads.

Yahoo's ad system gives the most prominent placement to the advertiser that bids the most per click. By contrast, Google weighs additional factors, such as the frequency with which consumers click on each ad, to determine the order the ads are displayed.

Google's approach generates more revenue per search, since the most popular ads for any query appear more prominently, increasing the likelihood consumers will click on them. Compared with Yahoo, Google brings in roughly 45% more revenue on average for each search query it handles, according to Majestic Research Corp., a research firm based in New York. Yahoo executives concede the structure of its search-ad system has cost the company hundreds of millions of dollars in revenue.

Yahoo has consulted with economics professors and its own research staff for a project called "Panama," the company's ongoing attempt to revamp its search-ad system. Mr. Raghavan vows that economists will be involved alongside Yahoo computer programmers in building future services, so as to head off future mistakes.

The use of incentives is one of the research areas that interests Mr. Schwarz, the economist who begins at Yahoo next month. He's currently writing a research paper about economic markets where one side of a potential transaction is eager to profess its preference for the other side, as is often the case with hiring and online dating. The unequal relationship makes such markets complicated and unpredictable. The research has clear applications for Yahoo Personals.

Mr. Schwarz also co-wrote a paper examining search-ad auctions. It concluded that Yahoo's system doesn't encourage people to bid what they think the ad is worth to their business, something that caught the attention of Yahoo executives. They expect Mr. Schwarz to work on future versions of their ad systems, a task he has already embraced.

"This is the type of problem economists have been studying for the past 60 years," Mr. Schwarz says.<\lj-cut>

  • Post a new comment


    Anonymous comments are disabled in this journal

    default userpic

    Your reply will be screened

    Your IP address will be recorded