Just one in every 20 Facebook photos is shared. Most Twitter hashtags vanish into oblivion. Researchers want to understand the exceptions. Can you predict what content will go viral? That could be handy in many contexts—marketing, elections, revolutions.
This week two new papers are proclaiming advances in the science of virality.
“We have a method that allows us to predict the future on Twitter,” says James H. Fowler, a professor of medical genetics and political science at University of California at San Diego.
In a paper published on Wednesday, Mr. Fowler and four co-authors describe a technique for mining Twitter to forecast, about a week in advance, which hashtags will become popular. The method is to compare a randomly chosen group of tweeters against another group of people who are more central in the Twitter network. When the better-connected people start using a hashtag more than the randomly chosen people, that flags something that is going viral.
Mr. Fowler’s team identified the “central” group through an old social-science concept known as the “friendship paradox.” The Twitter version of it goes like this: Take a random person on the social network, and count his or her followers. Then choose one of those followers at random, and count how many followers that person has. Most of the time, the followers have more followers than the first person selected. Mr. Fowler calls such people the “life of the party.”
“They’re going to get whatever is spreading through the network first,” he says. “Because the numbers are so large"—50,000 people in each group—"it’s easy to detect” when the more central people are using a hashtag that’s not being used by the random people.
Easy with the historical Twitter data used in this analysis, anyway. A skeptic would want to see how well the method performed live. “What we’d like to see is a website where, in real time, we’re using this method to say, OK, this is what we think is coming down the pike,” Mr. Fowler says. “But we have not built that yet.”
In the other new paper, presented this week at the International World Wide Web Conference, in Seoul, South Korea, researchers mined Facebook data to tackle a related problem: forecasting which photos will go viral. The best predictor of a photo’s popularity was the speed with which it was shared, according to researchers at Stanford University, Cornell University, and Facebook. Even so, the researchers found “no simple trick to ensure widespread sharing,” according to a Stanford news release.
“Even if you have the best cat picture ever, it could work for your network but not for my boring academic friends,” Jure Leskovec, an assistant professor of computer science at Stanford, was quoted as saying. “You have to understand your network.”
