Can I have your attention!HP Labs pilots i-catcher, a service that automatically maximizes the viewership of media Web sites
By Simon Firth
Not so long ago, we had relatively few places to go for news. We could turn to a city newspaper or two, a few radio and TV networks and some weekly news magazines.
But these days, says HP Labs Senior Fellow Bernardo Huberman, “the scarce thing is not information. It’s attention.”
In an era that’s added the entire Internet to the multi-channel offerings of cable and satellite TV, consumers are spoiled for choice. And news editors are being pressured as never before to maximize the number of people reading, hearing or viewing the material they offer.
HP Labs Senior Fellow
Bernardo Huberman and
researcher Fang Wu.
Behind i-catcher, says Huberman, is a complex algorithm that balances novelty with popularity. “We all attend to things online because they are new or popular,” he explains. “Eventually attention fades, and we attend to other things that are new or popular.”
By factoring that decay into account, i-catcher can help editors avoid keeping an item on a page long after interest in it has waned. Similarly, it can give new items a chance to shine. But if they’re not proving interesting to anyone, i-catcher will suggest moving them lower down the page or removing them altogether much more quickly than conventional editorial practices would dictate.
In maximizing the potential audience for every item, i-catcher maximizes the potential of the entire homepage. And in doing that, Huberman notes, “it maximizes the use of two limited resources: the total visual space available to the content provider and the total attention that a group of users can pay to it.”
The demise of human editors?
One way to think of the service is that it rationalizes the traditionally intuitive online editorial process.
So does that mean that HP’s i-catcher might come to replace human editors – as the Economist magazine suggested when it profiled Huberman and Wu’s research in a recent article ?
Not necessarily, says Huberman. “An editor may still want to prioritize some things over others,” he notes. At the same time, though, whenever editors have the urge to offer a higher profile to an item that, for example, adds gravitas to the site but which few will read, Huberman suggests they may also want to know what i-catcher recommends.
After publishing the theoretical work behind i-catcher last year, Huberman and Wu began a pilot implementation of the service three months ago with a major online publisher in Scandinavia which posts links to hundreds of news stories on its homepage every day.
While the final results of the pilot are not in, i-catcher is already making suggestions for page placement of specific items that are dramatically different from recommendations based on either popularity or novelty – the two most common ways of ordering online content today.
In the future i-catcher might be employed not only by media organizations, but by groups with retail Web sites like HP itself.
The basic algorithm would need to be tweaked, Huberman notes. Technology products are novel for weeks rather than a few hours, for example. And HP might want to use criteria like available inventory or profit margins in deciding which items to favor. “But it will still tell us what items we need to move up the page to maximize the attention these items receive in total,” he says.
“The whole idea here is to look at the economics of attention,” adds Huberman.
That's been a major focus of HP's Social Computing Lab for some time. Huberman and colleague Gabor Szabo have a related project that can predict the future popularity of online content just from the rate at which people are currently choosing to see it.
In January of 2009, along with colleague Daniel Romero, Huberman and Wu also published a study of social interactions at Twitter in the online peer-reviewed journal First Monday that proved hugely popular. The article, which revealed the “sparse and hidden network of connections underlying the “declared” set of friends and followers,” has already been downloaded some 25,000 times.