The New York Times has been talking for a while about how it wants to create a more personalized, engagement-oriented user experience. And quietly, late last week, the paper took another step in that direction: It launched nytimes.com/recommendations, a customized story-recommendation page. Most notable about the page — beyond, of course, the awesomely mohawked-meets-buzzcut default profile pic, the same one you’ll find on Times People — is the “What you’ve read” box on the page’s right, which displays, in an almost sports-stats-like way, how many, and what kind of, Times articles you’ve consumed over the past month. The totals are broken down both by section — Business, Magazine, U.S., etc. — and, more broadly, by topic. (My most popular include “Computers and the Internet,” “News and News Media,” and, um, “Murders and Attempted Murders” — the last of which I’m really, really hoping has to do with my following of the events in Tucson…)
Particularly during a time of transition at the Times — the paper’s paywall is set to rise any day now — the Recommendations engine is an intriguing new feature. I spoke with Marc Frons, the Times’ CTO for digital operations, to find out more about the engine — and how it fits into the Times’ broader strategy for user engagement.
“The whole idea is to expose our readers to as much of our great journalism as we can,” Frons told me. On the web, it can be hard to find the things you like — not to mention the things you don’t know you’d like until you like them. The new Recommendation engine, Frons says, “allows us to expose content to our readers that they wouldn’t see any other way.” And it allows the news organization, more broadly, “to establish a more personal connection between what we do online and what our readers do online.”
While a lot of the rec engines out there are framed around content that users have read previously, “what we try to do is look at people’s patterns, and how they move around the site, and what sorts of different things they might look at,” Frons notes. The engine tries to accommodate the complex dynamics of usage and movement as people navigate a through a news experience. The Recommendations page displays a range of stories — stories that are connected, in particular, “by your various interests,” Frons says, “not just what you looked at last.” (That focus on pattern is the reason the Times built its own engine, in fact, rather than teaming up with an existing personalization platform.)
The bonus for the user (and, I’d add, for the paper that wants to encourage user loyalty): The more Times articles you read, the more relevant, ostensibly, the recommendations will be.
But while the system is educating the Times about your interests, it’s also educating you about them. As Frons puts it: “It’s kind of like taking one of those personality tests where it tells you things about yourself that are only obvious in retrospect.” And the numbers — X stories about “World,” X stories about “Entertainment,” etc. — are stark. They quantify news consumption in a way that news consumption is generally not quantified. It’ll be interesting to see, from a user-interface (and, really, a user psychology) perspective, whether that running tally of stories read, and categories and topics followed, will affect what kind of news people choose to consume. Will the Times’ all-seeing eye — “we’re not judging, or anything, but you seem to read a lot about murder” — encourage me to read more about, you know, kittens? Or at least about politics? And will the “What you’ve read” totals double as a kind of graded assessment, encouraging us to do better — ie, read more articles — next month?
The recommendations project has been in the works, conceptually, for about nine months, Frons estimates — and in actual development for the last three or four. Derek Gottfrid started it off (and “we couldn’t have done it without Derek,” Frons notes); once Gottfrid left the Times to work at Tumblr, the paper’s software group took over the engine’s development. The team rolled out the engine Thursday night, in a rare-for-the-Times soft launch. And that was in part because, particularly for something as personal as, you know, a personalized recommendation engine, user feedback is key to improving the engine’s functionality. As Frons puts it: “The recommendations cut both ways.”
Indeed, there’s a social aspect to this — if not the give-and-take of Facebook-like story sharing, the implicit interchange between the reader and the publication. I asked Frons whether the Recommendations engine is at all connected to News.me, the social news service designed in the Times Company’s R&D lab and being developed by Betaworks*. It’s not. “News.me is a separate strategy,” Frons told me. “It’s a blending of different technologies. It’s really not connected, except for the overarching idea that we want to give our readers as many choices as we can to look at our content, and read our content, on as many platforms as possible.”
For that matter, with its porous-paywall-esque “you’ve read X articles so far” framework, does the Recommendations engine have anything to do with the Times’ paywall? “Only tangentially, if at all,” Frons says. “Paywall or no paywall, our job, and the job of other news organizations, is to provide relevant news and information for our readers. To me, no matter what the model, the more people who read and are engaged with your website or your digital products, the better. So the recommendation engine just fits into our overall strategy of increasing user engagement.”
Nor, for that matter, does the recommendation algorithm use personal data to make its recommendations. “It really deson’t look at your demographic profile at all,” Frons says. Instead: “You are what you read.”
The plan is to roll out the engine more broadly over the next few weeks. “Right now, it’s a Recommendations page,” Frons notes. “But what we’re going to do eventually is make it more central to your experience on the site.” That could involve, for example, a “Recommended for You” tab next to the Most Emailed/Viewed/Blogged tabs on articles pages’ “Most Popular” boxes — a nice little nod to the twin virtues of the Internet news experience: personalization and serendipity.
As the engine rolls out, and as people respond to it, it will adapt. Right now, for example, the engine counts only articles that you consume on the web, not on mobile devices — but that’s only a temporary limitation. “It will eventually count across all devices,” Frons notes — iPhone, iPad, the mobile web, etc. — “because we know that people will want to read us on every device.” And that — again, the feedback loop — will mean more accurate recommendations. “The algorithm is evolving,” Frons notes. “We’ve played around with a lot of things, and it will continue to evolve.”
*This post initially said that News.me is a product the Times is developing in conjunction with Betaworks; I updated it with more precise wording to reflect the fact that News.me is a protoype that was designed in the Times Company’s R&D lab — and then sold to Betaworks in exchange for equity in Bit.ly. As part of the deal, a team of developers from the Times’ R&D lab worked at Bit.ly to help bring News.me to market.