You might not think an applied mathematician who does research in biology and has a PhD in theoretical physics would have much to offer a 163-year-old newspaper publisher, but Chris Wiggins, head of the data science team at the New York Times, told attendees at the Structure conference in San Francisco that machine learning can do much the same thing for media companies as it does for research biologists: namely, make sense of a whole pile of data.
In the case of the Times, that data is about things like what pages readers look at, how long they spend reading them, what they click on or read before and afterwards — and especially how that behavior relates to the paper’s advertising and the reader’s desire to sign up for or renew a subscription. And with the recent launch of two new mobile products, including the NYTNow app, the paper will have even more data on user behavior that it can play with and comb through for insights, Wiggins said.
One of the things that machine learning and data analysis can do, he said, is search through large datasets about user behavior in order to detect patterns that might indicate which direction the editorial or business side should take a particular product, whether it’s a web offering or a mobile app. Instead of deciding which features need to be focused on in advance, Wiggins added, “we can rely on the data to reveal which features we should be paying attention to,” because of the correlations in those patterns.
Although the New York Times hasn’t started doing so yet, Wiggins said he could see a point in the near future when the editorial side of the business might even start A/B testing headlines and other pieces of content — the way some digital-only publishers like BuzzFeed and Huffington Post do — provided that doing so doesn’t interfere with the commitment to editorial and journalistic quality that the paper is known for.
Photo by Jakub Mosur