A huge amount of content on Facebook Inc (NASDAQ:FB) is shared only a few times, but a few lucky posts are shared millions of times. Whether it’s text, video or photos, some content catches the eye of everyone, and leads to cascades where the number of reshares can reach millions.
Researchers have tried to find a way to predict which posts are likely to become more popular. The popularity of content depends on multiple factors, so some argue that it’s impossible to predict. But Stanford University professor Justin Cheng, along with researchers at Cornell University, has demonstrated that many characteristics of a cascade can be predicted with a very high degree of accuracy.
At this year's SALT New York conference, Jean Hynes, the CEO of Wellington Management, took to the stage to discuss the role of active management in today's investment environment. Hynes succeeded Brendan Swords as the CEO of Wellington at the end of June after nearly 30 years at the firm. Wellington is one of the Read More
Researchers tracked 150,000 photos on Facebook for about a month
According to the MIT Technology Review, Cheng and his team analyzed the way photos were shared on Facebook Inc (NASDAQ:FB) over a period of 28 days after their initial upload in June 2013. They tracked 150,000 photos that were together reshared more than 9 million times. They collected data on which people reshared each of the photos, and at what time. It allowed them to figure out the networks through which these photos were reshared.
Cheng and his team started with a photograph which has been reshared a certain number of times, say X. Then they determined the likelihood of this photo being shared twice as many times. In simple terms, they just tried to predict if the cascade will double in size. That’s a useful method because the cascade distribution size follows certain power laws. The law ensures that, in a given size of cascades, half will fail to double in size while the other half will easily more than double.
Cheng and his team used this dataset to create an algorithm to figure out features of cascades to make them predictable. The features include the number of followers the initial poster has, the speed of cascade formation, the type of image (outdoors or close-up, with or without a caption, etc.), and the shape of the cascade that forms. After training their algorithm, the scientists used it on other Facebook Inc (NASDAQ:FB) posts to know if it can make predictions about other cascades.
Researchers began with those photos that had been shares only five times. So, the algorithm had to predict whether they will be reshared more than 10 times. The scientists found it surprisingly predictable. The algorithm predicted it with accuracy of 0.795. They found that some features of the cascade are much better predictors than others.
The study revealed that the temporal performance of a cascade is the best indicator. That means if a cascade spreads quickly in the beginning, it is highly likely to spread more. The next important factor is the topic mentioned in the photo or video caption.
Facebook Inc (NASDAQ:FB) shares jumped 1.57% to $65.11 at 12:48 PM EDT.