Algorithms are a crucial part of how the news reaches you in a digital world. But we know many people find them opaque and controversial. Filter Bubble is an expression coined to capture the way an algorithm can measure what you like and just feed you more and more of that until all you get is one perspective.
We want to raise the curtain and explain how we use an algorithm at NPR One.
If you run a Google search on the word "algorithm" it will be explained as "a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer." There's nothing inherent to an algorithm that makes it good or bad. It is just a set of procedures. The question to ask is what are the people who programmed the algorithm trying to accomplish in using it?
In building NPR One, our commitment was to maintain the core values of NPR's journalism while taking advantage of the abilities of a digital platform to customize a listening experience for you. When you listen to a radio show, the producer of the show has made editorial decisions about what you hear. When you listen to NPR One, the algorithm is an editorial tool we are using to ensure you get the appropriate pieces and recommendations at the right time.
So here is what you can expect from NPR One's personalization algorithm:
NPR One will always give you the lead story of the day and the important stories that our editors have chosen. Experienced news producers and editors make the decisions about what stories need to be offered to you to ensure that you are well informed and that you get a couple of surprise selections that might be things you didn't know you were interested in. We know you value getting both the information you need along with the serendipity of discovering something new—that is the heart and soul of NPR. We've set up our personalization algorithm to ensure this is a big part of the NPR One listening experience.
Stories are categorized in NPR One. Different types of stories act differently and allow for more or less personalization. Important news stories get presented to everyone with no personalization. As does another type of story which are ones that are hand chosen by editors to give you breadth and a little serendipity.
But there are other stories where it isn't essential that everyone hears them, stories that are basically "water cooler" sorts of stories or stories that are more particular to one person's tastes and interests. With stories like this we feel the algorithm should support your interests. For instance, NPR One can learn that you don't really like long interviews with novelists, lots of sports stories, or music reviews. However, even in those cases, our personalization algorithm is designed to periodically double check your preferences by offering you something it thinks you'd prefer not to get, just to make sure. And if something in one of these categories is particularly newsworthy, our editors can make sure it isn't subject to personalization so you'll be offered the story. We're also working on improvements to the personalization algorithm to learn even more about the topics you favor. This will help you dive deep into topics you are particularly interested in, while still ensuring you get a clear view of the broader world. Our goal is to help you break filter bubbles.
The other place where we rely on our personalization algorithm is for determining what podcasts are recommended for you. Most podcasts are not where you turn for your essential news. So, in this area, we feel we can make heavier use of an algorithm to make choices about what to play and recommend. Our editorial curators still have the ability to decide to present a certain particularly noteworthy podcast far and wide to our listeners, but we feel that podcasts are another area where a personalization algorithm has real power to improve your listening experience without compromising editorial values.
Ironically, while algorithms have been blamed for fostering "bubbles" where you see, read, and hear only the things that reflect your world view, we can use the NPR One algorithm to ensure that you actually get editorial balance. It is not uncommon for our on-air broadcasts to feature one point of view one day and the other point of view on another day. Maybe you will hear both if you listen on-air, or maybe you won't. It will depend on whether you tuned in at the right time. With NPR One, we can have our personalization algorithm make sure you get both. It also gives us the ability to make sure you get a follow up story when something changes or a correction is made. This is an example of how the use of an algorithm can actually be more editorially responsible rather than less so.
Ultimately, an algorithm is just another editorial tool. Most of the work the NPR One personalization algorithm does is benign housekeeping such as making sure you are getting the most recent important stories, that you aren't hearing the same story twice, and that you get stories from your local station not some other one. But algorithms can also be setup to maximize whatever it is that the website, social media outlet, or news organization values most—in NPR One's case, that is stories that start conversations, increase understanding, enrich lives, and enliven minds.
Tamar Charney is the Managing Editor for NPR One; Thomas Hjelm is NPR's Chief Digital Officer and Michael Oreskes is NPR's Senior Vice President of News and Editorial Director.
Copyright 2022 NPR. To see more, visit https://www.npr.org.
