Stark white and minimally designed, the new NPR One app looks like a paradigm of technology. But surprisingly, the app isn't powered by algorithms, filters, or other pseudo-intelligence—it's still good old human editor curation on the backend.
"For us, the algorithm that programs the app is very importantly focused on the human curation part of it," says NPR VP of digital media Zach Brand. "A lot of people tend to think of it in terms of machine learning—which is a portion as well—but we have dedicated staff making sure that the most important stories are populated from the outset that represent the best experience right at the first moment. As we get to know the listener, it then tailors even more to them."
For Brand's team, using human curation as a backstop ensures the app doesn't screw up the most vital interaction: the first-time user experience. Usually, algorithmically personalized apps struggle to meet a user's expectations until the app has enough behavioral data to make inferences about what the user likes. In the NPR app, the first time a new user opens the app all the stories they see (or hear) were picked by human editors. From there the app begins to take cues based on some of the indicators and transitions to more machine learning.
"We absolutely can use and do use machine learning in here, but [we] are very true to the experience of public radio," says Brand. NPR didn’t want the app to be a filter bubble for listeners. It didn’t want to artificially narrow the scope of stories people were hearing because they were selecting topics and categories. Another thing it didn’t want was to only be a destination for the most popular stories. There’s a mix of local and national stories along with popular and obscure ones.
The actual app is simple looking and was built as a minimal viable product (MVP). Front and center is a play/pause button, 15-second-rewind button, and a skip button. Swiping left in the main area shows the upcoming story segments while swiping right shows a history of what you’ve already listened to. There’s a share button, an "interesting" button, and search, but there’s not much else—not even settings.
"One of the interesting things we saw [while beta testing] was that people didn’t want to like or dislike news stories, because it’s pretty hard to thumbs-up a genocide story," says senior director of digital products at NPR Joel Sucherman. "We’ve seen the 'mark interesting' as the most powerful indicator to reliably know someone’s interest."
Another on-demand audio app, Swell, was just bought by Apple a little more than a day before NPR One launched. Swell collects different parts of podcasts and audio content and ties them together, also with a swiping gesture. The app swings heavily in the opposite direction of NPR’s app—it's fully automated to learn based on user preferences and skipping signals.
The Tribune Company also released its Newsbeat app earlier this year, which works a little bit differently. The talk radio app takes text articles from websites and with a text-to-speech process, turns the writing into audio. Newsbeat also employs human talent in a studio to constantly read and record the most popular stories.
[Image: Flickr user Todd Huffman]