"Users can tell Zite a lot about the news articles they know they like to read: topics, authors, publications. In order to help them expand their minds, though, we build serendipity into our algorithm. Even though we base our suggestions on relevant factors--what kinds of articles users are sharing on social-media sites, what others like them are reading--we still often show people stories we're not 100% sure they'll like. That can lead to an article popping up about aardvarks migrating to Washington, D.C., which happened in my Zite, or to us getting emails complaining that someone's politics section is too conservative or too liberal. I always halfheartedly joke that if people only ever 'thumbs-up' the articles we show them, then we've failed, because we're not letting them experience anything new."
FAST COMPANY: What's challenging about discovery versus search?
MARK JOHNSON: When you search for a term on the web--for example, "Walmart"--you're going to get Walmart.com, because search engines are optimized to figure out how to do that. But if you want interesting articles about Walmart, there's no term that you can type into a search engine to find that. When I became involved with Zite two and a half years ago, Zite had all this wonderful technology that was able to deliver people these highly personalized results, but the web wasn't really the right platform. It was hard to get people to install the plug-in.
How did you settle on an iPad magazine as your product?
The iPad really filled a hole in the market that no other device had. This is something that people can plop on their couch with and read, rather than having a lot of books and magazines and newspapers. That's really the context in which news discovery becomes most interesting, when you're in a very comfortable situation. It's harder to imagine a family of four sitting with their laptops on a couch consuming news. We realized we could create a service different from anyone else's.
How do you go about wading through the entire Internet for Zite content?
It would be impossible to manually curate the thousands of topics on Zite. So first, we look for things that have some sort of social velocity--what's being talked about on Twitter or other social mediums like Delicious. That gives us an idea of what people are talking about and what's interesting to them. From that, we get millions of documents a day that are potentially going to show up in someone's Zite. Then we look at all these documents and do an analysis: What's the topic about? What's important in this article? How is it written? How long or short is it? All of these can factor into the kinds of documents you like to read. Finally, we apply a collaborative filtering step, which essentially says people like you probably like things that you're going to like as well. We've got a lot of secret sauce there in order to figure out what kinds of ratings we give articles, and we take all sorts of signals from the social web to make this happen. Zite wouldn't have been possible prior to the advent of people sharing so much because we rely very heavily on such signals.
Where do you see the future of personalized news discovery going?
Part of the problem with the news today and news discovery is we've been so satisfied with good enough for so long that we haven't opened our eyes to the possibility of what news discovery can be. There's lots of great content being produced every day. The reality is you just don't see most of it because it's impossible to find. I hope we're on the forefront of changing peoples' perceptions about that. The number one comment we get from Zite users is, "Wow, you've introduced me to things that I never would have otherwise read."