What makes it go viral?
Ever since the very first cute kitten video (or whatever else it was that went viral), individuals and marketers have been looking to post content online that attracts millions of views. Now, researchers from Stanford, Facebook and Cornell say they have developed a system, which can predict viral events up to 88 percent of the time.
The key to the research was finding patterns in "cascades," which describe videos or photos that have been shared multiple times. The problem with researching cascades, according to the researchers, was the infrequency with which they occurred. Looking at photos on Facebook, the researchers discovered that only one in 20 photos posted on the site get shared even once. Only one in 4,000 gets shared more than 500 times.
The researchers analyzed 150,000 photos from Facebook that had been shared at least five times. The names of individuals and other identifiers were removed to protect privacy. A preliminary analysis showed that there was a 50-50 chance of shares doubling at any point during a cascade.
Next, the researchers looked for variables such as the rate and speed at which photos were shared, how many networks they were posted in, etc. to identify patterns.
The best indicator of a photos "viralbility" was speed of sharing. Using this variable, researchers could correctly predict cascades 78 percent of the time.
The next best predictive factor was the structure of the cascade — how items are spread among friend networks or interest groups. This proved to be 67 percent accurate in predicting doubling when used alone.
"Even if you have the best cat picture ever, it could work for your network, but not for my boring academic friends," Jure Leskovec, assistant professor of computer science at Stanford, told Stanford News. "You have to understand your network."
After looking at several predictive criteria, the scientists were able to predict doubling almost 80 percent of the time. As photos were shared more often, the accuracy rate went up to 88 percent.
The research team on the project included Justin Cheng (Stanford), Lada Adamic and P. Alex Dow (Facebook), and Jon Kleinberg (Cornell) in addition to Prof. Leskovec. Their findings will be presented at International World Wide Web Conference.
- Stanford computer scientists learn to predict which photos will go viral on Facebook – Stanford News
- Scientists Predict Viral Hits With 80% Accuracy – Springwise
- Viralbility – Urban Dictionary
- Branded Viral Videos: The Secret Marketing Weapon – Mashable
- Why videos go viral – TED
Do you see value to marketers in the findings of the viral events study? How important is it for marketers to get a handle on viralbility for social marketing? What do you think are the keys to a photo or video going viral?