Personalization Done Right — The Harmony of Science and Human Insight at Yahoo!

At last year’s Yahoo! Labs Summer Science Week we celebrated the five-year anniversary of Yahoo! Labs. This year at Summer Science Week 2011, we’re celebrating another momentous milestone – the popular “Today Module” (the set of revolving news stories you can see at the center of the page) on the Yahoo! Homepage has now surpassed an average of more than 1 billion clicks in the U.S. per month throughout 2011.

While the sheer scale of reaching 1 billion clicks per month is noteworthy, from a scientist’s perspective, what happens behind the scenes is just as impressive. The “Today Module” is powered by a suite of technologies we call the Content Optimization and Relevance Engine (C.O.R.E.) that had its genesis three years ago in Yahoo! Labs. For the scientists at Yahoo! Labs, the C.O.R.E. project is the kind of complex, challenging work that makes it fascinating to work at Yahoo!.

At the Heart of C.O.R.E.

The goal of the C.O.R.E. project was to marry machine intelligence with the human intelligence of Yahoo!’s editorial vision and voice -- and to do it at Yahoo!’s massive scale of 680+ million users. Having developed tremendous algorithmic expertise in matching Web content to search queries on Yahoo! Search, we were eager to apply similar techniques to the Yahoo! Homepage.

As the graphic below illustrates, the system our scientists developed in close collaboration with our product, engineering, and world-class editorial teams has been an amazing success. C.O.R.E. enables the “Today Module” to deliver 45,000 different, personalized variations of news and media content to Yahoo!’s audience every 5 minutes!

Personalization Done Right

We appreciate that so many people visiting Yahoo! love these personalized experiences, but there’s more to this story. We see C.O.R.E. and the specific type of personalization it delivers as having a major impact not only on Yahoo!, but also on the future of digital media and online journalism in general.  Let me explain.

With all of the content and media available to us on the Web today (and it’s a lot, Yahoo! alone adds 50 TB of data every day), personalization makes a lot of sense – it connects people to what they care about most. But there are worthwhile debates in our industry about what many experts argue is the risk of “over-personalizing” the Web to the point where everyone lives in their own personal “Filter Bubble”.

And that’s where C.O.R.E. comes into play. C.O.R.E. was built to avoid this “over-personalizing” in two key ways.

First, C.O.R.E. estimates the popularity of stories in multiple ways. For example, we calculate how much we think an article may appeal to the entire U.S. population, as well as how much we think it may play with smaller groups of users (like women or men, or men in San Francisco), and all the way down to an individual user.  Then we vary the weight of those different estimates to reflect the diverse interests of all our users. So people often see articles that are popular with other Yahoo! users – even though these articles may not align precisely with their own personal history or preferences.  The result is an experience that feels personalized and relevant, but is also serendipitous, diverse and fresh.

Second, C.O.R.E. blends the judgment of our editorial teams with our data-driven ranking technology. That way we optimize based on real-time and historic data regarding a consumer’s location, interest and intent, while also empowering editors to make informed decisions about prioritizing editorial coverage and vital news events as they happen.

This interaction produces personalization done right because it balances the power of personalization technology with humility in the face of a world where what’s important and “relevant” is inherently dynamic and unpredictable.

If you’re curious about how this matchmaking between science and editorial voice comes together, here’s a behind the scenes video from the Yahoo! Homepage team:

As the results show, this form of personalization gives consumers stories that match their interests, keeps them informed about the world at-large, and constantly surprises them with fun, related content to discover.

Yahoo! Labs – A Year of Milestones

Personalization is just one of many interesting technical opportunities we’re tackling at Yahoo! during Science Week. For a full download on what’s happening at Yahoo! Labs, check out our website, where you’ll find a wealth of information on the banner year 2011 has been for Yahoo! Labs, which included the opening of Yahoo! Labs Beijing, the five-year anniversary of Yahoo! Labs Barcelona, the launch of Yahoo! AdLabs, “Best Paper” and “Best Poster” at WSDM 2011, and the “Best Student Paper” award at ACM SIGIR 2011. You can also keep track of all the innovative projects from Yahoo! Labs by following our Twitter feed.

Finally, on this very blog later in the week, we’ll be showcasing video interviews with some of our scientists who will be sharing their thoughts on personalization and where the science of the Web is heading. Be sure to check back in tomorrow.

 

Prabhakar Raghavan

Yahoo! Chief Strategy Officer, Executive Vice President, and Head of Yahoo! Labs