Before declaring the least-favoured – but most-covered – U.S. president-elect in recent history Person of the Year, the editors behind New York City-based Time magazine consulted an unlikely source for advice.
“What they told us when we first started talking to them is that [choosing Person of the Year] was getting very repetitive,” André König, co-founder and head of strategy and revenue for Opentopic Inc., which collaborated with Time on its newest experiment, tells ITBusiness.ca. “Castro hadn’t died, Donald Trump wasn’t elected yet, and thinking about the stories they could write about whoever they picked was a real challenge for them. So they were looking for a way to make it more interesting.”
While the people’s choice for the person who wielded the most international influence this year – Indian Prime Minister Narendra Modi – did not win, König says that Time’s editors used a combination of machine learning and reader feedback to narrow down the candidates.
They started by approaching IBM, he says, thinking they could use the Watson platform to analyze the year’s media coverage in an attempt to both narrow down the field and come up with ideas for how to approach writing about whoever they chose.
IBM, in turn, contacted Opentopic, which specializes in target audience insights and has been an IBM partner for around three years.
“Publishers in general – and we work with a bunch of them: the Economist, [Time magazine parent company] Time Inc., Conde Nast – have a couple of big structural problems,” König says. “Finding your target audiences, engaging them, and making sure they come back and, ultimately, subscribe to your publication is getting more and more challenging.”
“One way that Time Inc. is hoping to address the challenge of engaging audiences is finding better, more data-driven ways to create editorial content,” he says.
Typically when selecting the magazine’s Person of the Year, Time’s editors will discuss who they believe most warrants the designation during a two- or three-week period anyway, König says, so inviting the public to join was only a short hop away.
But first, they gave Opendata a list of “a few hundred names,” he says, “and we analyzed them all – content, media mentions, publications – for 2016.”
For each person studied, Opendata gathered information from across social media channels, online publications, bloggers, RSS feeds, and webpages before running it through multiple Watson-driven APIs (advanced program interfaces), including sentiment analysis, personality analysis, concept extraction, relationship extraction, and geolocation. It then presented the data in what König called a “user-friendly” dashboard-based format that allowed Time’s editors to compare and contrast the results.
The Watson platform was essential on two fronts, König says: The level of analysis Time’s editors were looking for required a great deal of server power and technical flexibility, since Opendata was analyzing multiple types of content and millions of pages.
“Everything we do at Opentopic is hosted on IBM Bluemix SoftLayer, which is a very flexible server infrastructure that allows you to scale up very easily and control what you do,” König explains. “It is also integrated with all of the IBM Watson APIs, so it basically allows you to turn any kind Watson API that you want on or off within a couple of hours.”
“It’s very flexible, which was very important for this project, since it had an iterative discovery process,” he continues. “We’d provide a set of analyses, the editors would look at it and come back with some feedback – stuff they like, stuff they don’t think apply – and we’d move forward from there. The IBM Bluemix SoftLayer infrastructure was critical to that.”
In the end, Opencontent analyzed tens of millions of pages across multiple parameters, providing a level of insight and decision-making that humans couldn’t, König says.
“For Time, it’s hopefully a better way to pick their person of the year,” he says. “It’s also a good story in itself.”