Although the bulk of content on Facebook is shared only a few times, some can be reshared millions of times, and researchers at Stanford University, Cornell University, and Facebook have demonstrated that various traits of a cascade can be accurately predicted and applied to making successful assessments about cascades' future behavior once they have begun.
An analysis of how photos were shared on Facebook over a four-week period following their first upload revealed which people or nodes reshared each photo and at what time, permitting precise reconstruction of the networks through which the reshares transpired. By starting with a photo that has been reshared a certain number of times, and then determining the probability that it will be shared twice as many times, the researchers establish a specific kind of power law the distribution of cascade size follows. Therefore, a random guess will yield the right answer about 50 percent of the time.
The researchers use a portion of their data to train a machine-learning algorithm to seek cascade features that inject predictability, such as image type, the original poster's number of followers, the cascade's shape, and the speed at which it occurs. They found that their method achieves a classification accuracy of 0.795, versus random guessing's 0.5.
From Technology Review
View Full Article
Abstracts Copyright © 2014 Information Inc., Bethesda, Maryland, USA
No entries found