HP Studies Online Attention and Learns to Predict Virality

Unearthing Early Patterns in Videos and News Stories Can Help Predict Ad Revenue, Optimize Web Content

By Published on .

NEW YORK (AdAge.com) -- In a world where information is exploding yet attention span is static, anything that can help predict what kind of content will attract eyeballs and cause virality is valuable. And research from HP's Social Computing Lab suggests it can be done.

Bernardo Huberman
Bernardo Huberman
"Information has collapsed in value because there's just so much available ... but the scarce commodity is attention," said Bernardo Huberman, senior fellow-director of the social-computing lab at HP.

The Social Computing Lab comprises 14 people whose backgrounds range from computer science to behavioral economics, and the group deals in the intersection of social behavior and information technology. And recently, it has focused on understanding the dissemination of content, finding influencers and how to induce virality and crowdsourcing.

For one study -- which aimed to come up with a way to predict the popularity of online content -- Mr. Huberman and his colleague, Gabor Szabo, studied 7,000 YouTube videos and 60 million stories on news-sharing site Digg. The two uncovered a trajectory or curve that describes how attention is allocated among all content over the life cycle of a typical YouTube video or Digg story.

In the case of Digg, looking at the performance of the first two hours of a story allowed them to forecast the story's popularity 30 days out. In the case of YouTube, looking at a video's data for 10 days could allow it to do the same for that video.

'Much like radio activity'
"If you look at the average -- how many hits-per-unit-time a video gets -- and you look, over time, when the video first appears, it decays and it decays in a particular way, much like radio activity has a half life," Mr. Huberman said. "And knowing those things allows us ... to predict how many hits it'll get in the future."

The knowledge is useful to HP's own marketing efforts: It lets the company's marketers understand whether something that's buried deep in one of its websites is exhibiting potentially viral behavior early on, so they can move it up on the website, supplanting a link that's starting to decay. But it could also help predict ad revenue for online videos or pieces of content. The study was also the basis for a product the company is testing -- an algorithm-based content management system called i-Catcher, which reorganizes content on a website based on people's attention to it.

The lab launched nine years ago, when HP recruited Mr. Huberman away from Xerox, and it creates products and services that can be used internally and sold externally. Its CloudPrint invention lets people print from any phone without installing drivers, and is now a product at HP. Internally, the lab deployed Watercooler, a tool it created to tap into the social-media conversation at HP and then disseminate the popular topics of conversation to employees. And a few years ago, it introduced Brain, which pays small groups of people based on how well they can predict the future, such as how many products will be sold or the likelihood a product will be approved. Most recently it introduced HP Gloe, a crowd-sourcing-based local-information application that launched last week on the mobile platform Android.

From a marketing standpoint, its discoveries around how opinions form and propagate through the web have proven valuable. In one study, it found people are inclined to offer differing opinions than the most obvious prevailing sentiments when writing product reviews. Its study of the dynamics of viral marketing dispelled several myths associated with viral -- such as viral things propagate forever -- and also demonstrated limits to the effectiveness of the highly connected influencer.

Said Mr. Huberman, "Understanding social attention -- not just how the mind of a single person works, but how millions of people suddenly go and download a particular video or notice a particular ad or buy a product -- is the essence of attention and therefore what advertising and marketing are all about."

In this article:
Most Popular