Predicting Popularity of Facebook Posts
https://www.technologyreview.com/s/525821/the-curious-nature-of-sharing-cascades-on-facebook/
Social networks have always been especially insightful for gaining data about networks because you can observe and analyze networks principles in action, such as information cascades. This article from Technology Review, which includes research from Cornell University, discusses information cascades in the context of post sharing on Facebook. More specifically, the article asserts that it is possible to predict the popularity of posts moments after they are shared by studying the principles of cascades.
In the course, we’ve learned that information cascades have proven to be a reason for many fads and one of them can be the sheer amount of people that choose to share a post. We can create an analogy between the example given in class, guessing the marble majority, to sharing a post. The first person and even second who sees a post may make their own decision on whether or not to share a post. With more public information as they see others sharing the post, this may lead them to rationalize that sharing the post is in fact, a good idea. If so many people before myself already shared the post, then they must have some information that I don’t have, right? In the article, the researchers were able to determine a machine learning algorithm that predicts whether the cascade (the network of people who shared a given post) doubles in size. They were able to find that the more quickly a post spreads out with to begin with, the more likely it is to spread in total.
As we’ve learned about information cascades, people often base their decisions on what others have already decided, inferring that other people must have some information that they do not. More information also helps the researchers from the article make their predictions on the popularity of Facebook posts. Justin Cheng, a researcher at Stanford University, states that “more information is always better: the greater the number of observed reshares, the better the prediction”. I see a lot of applications to this research, which predicts the cascade popularity of posts with a classification accuracy of 0.795. Perhaps it can be used by advertisers to predict how popular their ads are and will be so they know where to channel their funds into. In another case, social media and news outlets can use this research to predict the spread of fake news and direct efforts to terminating the posts that have the biggest popularity potential.