If you’ve flitted across the Yahoo home page recently, you know how addictive its Today module can be. That’s the little box at the top of the page (as in the image above) containing four news stories, including at least one you usually can’t help but click on.
And Yahoo says that’s entirely intentional, the result of a lot of hard work spent trying to figure out how to serve up news that you, yes you, will find irresistible.
The company started work on a powerful personalization algorithm four years ago. Now it’s paying dividends. The system generates 45,000 totally unique versions of the Today module every five minutes. (All five screenshots in this post were taken within minutes of each other, using different Yahoo accounts.)
Yahoo is now in the process of rolling out the algorithm–it’s nicknamed CORE (for Content Optimization and Relevance Engine)–to its other media properties. It hopes CORE’s results will entice visitors to gobble up ever-more content in the Yahoo ecosystem, and in the process, turn it into the “premier media company” that its executives like to claim it already is.
They might be on to something. The Yahoo home page is the most visited of any page in the U.S. containing programmed content. Every day, 35 million unique visitors give it a look-see, and 110 million stop by every month. The increase in number of clicks in the Today box between this year and last–167 million–is the equivalent of adding an entire LATimes.com to Yahoo’s content ecosystem.
So how does the personalization strategy work? Yahoo gave Fast Company a peek behind the curtain, and it turns out that the secret sauce is only partially due to the work the algorithm does on its own. The other impact it’s having is on Yahoo’s Front Page editors, who are getting savvier about what will appeal to readers, in large part due to the insights they’re gaining from the algorithm.
Here’s how it works.
CORE’s job is to figure out which stories, which come from Yahoo’s own content producers and external sites and publications, are going to play well with which specific audiences. Yahoo generates a profile for each user based on information they’ve entered about themselves, like gender and age (if they’re a registered Yahoo user), the places they’ve visited when they’ve come to Yahoo in the past, and the stories they’ve already seen during that particular visit. Based on that information, it’s up to CORE to figure out which of the 50-100 “packages” the editors have going at any one time will be most interesting to that particular visitor.
To test the packages, CORE grabs a portion of Yahoo visitors as they arrive on the site and uses them as a guinea pigs, tossing some of the new packages at them and seeing what attracts their interest. (Who gets lumped into that bucket is determined by a virtual “flip of the coin,” so that the pool is not always populated by the same people.)
CORE then uses the results of those tests to rank every package according to how well it will do with each visitor, based on their particular targeting attributes. So, for example, it might decide to serve up a particular package to women in the 35-44 age range in Peoria, for example, but not to men in that same age range in Peoria, or women in the 21-24 age range in Seattle.
“What we’re doing here is matchmaking,” Raghu Ramakrishnan, chief scientist for search and cloud platforms, tells Fast Company. “We know something about the user, about their context, and about the pool of articles we have. At the end of the day, it’s a matchmaking task.”
But the algorithm doesn’t work in a vacuum. It is also helping the editors who run the Front Page become smarter about how to put the packages together in the first place.
CORE can’t go hunting independently through the thousands of stories among Yahoo partners’ content, or the stories generated in-house, to find the pieces that might most interest Yahoo visitors. It totally relies on the Front Page editors to tell it what to look at, and that’s where some of the art comes in.
Some of the findings buck conventional wisdom. CORE has shown the team that boomers don’t just care about health and retirement; they also like music. Teens will click on parenting stories, as well as science and weather. And men actually are sometimes interested in fashion (which Lufkin chalks up, at least in part, to their trying to make sense of what their wives and daughters are wearing).
And while the average woman might not spend a lot of time following the ins and outs of football or baseball, some sports stories do capture their attention. The day that Fast Company visited the Front Page team, CORE predicted that a story about Derek Boogaard, the hockey player who died unexpectedly this spring, would do well with women–and indeed, according to the dashboard that Front Page editors monitor continuously, it did, perhaps because of the tragic nature of his death.
CORE is also helping Yahoo’s editors get a jump on upcoming stories. Months before the wedding of Prince William and Kate Middleton, the Front Page team noticed a rising interest in the bride’s sister, Pippa. So while it took the rest of the media a few days after the event to catch up to the frenzy surrounding the maid of honor (and her headline-grabbing bum), Yahoo’s editors were ready from the beginning.
But it’s not all fluff. The algorithm is also turbo-charging the team’s ability to tackle hard news. When Congresswoman Gabrielle Giffords was shot in January, for example, the editors tested a variety of angles to determine which one would grab the most interest. The winner, according to CORE? A story about the aide who tried to save her. It might not be what the New York Times would lead with, but it was the story that resonated most with the millions of people who landed on the Yahoo homepage that day.
“Our readers have a much wider range of interest than we’ve traditionally given them credit for,” Lufkin says. “If we can present them content that is compelling, there’s a really big opportunity” to grab their interest.
Having the editors at the wheel also means they can override the algorithm when important news is at stake. On the day Fast Company visited, President Obama was slated to give an important speech that evening on the draw down of troops in Afghanistan.
The algorithm predicted that the story on the speech would do miserably with Yahoo visitors. And indeed, according to the dashboard, it wasn’t getting many takers. But the editors still flipped the override switch, ruling that the story would be shown to all visitors to the home page at least once, irrespective of what the algorithm said. It was, and Yahoo willingly took the hit on clicks. Some stories, the editors say, everyone simply needs to see.
But, still, they monitored the story’s performance for any insights about what was resonating. “We want to learn how to do it better the next time,” Lufkin says, “not to sensationalize it, but to make it more relevant.”
As CORE moves out to other parts of the Yahoo network such as Yahoo News, where it’s headed next, the company hopes to reap a similar performance boost as it’s seen with the Today module.
Meanwhile, the company’s scientists are working on making the algorithm even smarter, determining what you’ll like not just based on simple variables, like your age and gender, but based on specific things you’ve expressed an interest in before, like “golf” or “Warren Buffet.”
And the human touch will conintue to play a part.
“From the beginning, we made the decision that we weren’t going to make everything entirely algorithmic,” Ramakrishnan says. “We need to leverage the editors and let them [use] the data we have to make smarter decisions in real time.”