MIT Sloan School of Management assistant professor Tauhid Zaman, Emily Fox at the University of Washington and Eric Bradlow at Wharton have combined Twitter and the future-predicting game Ouija to create “Twouija,” a model that they suggest can predict how popular a tweet will be.
How was the model developed?
The researchers found that during the 10 minutes after a tweet is sent, a famous tweeter and an unknown blogger might get roughly the same fraction of retweets (on average 50%) of the total number of retweets that tweet will ever get.
So even though someone with a massive Twitter following will get many more retweets than someone with a small following, the percentage of retweets both of their tweets get after a 10-minute period out of all retweets the tweets ever get will be very similar. That model is called a “log normal distribution.”
The Twouija model lets you select a Twitter user from a list of 50+ popular accounts – Barack Obama, Kim Kardashian, Fox News, CNET – and interact with a graph that plots the number of retweets of one of that user’s tweets over time. The researchers’ prediction, based on that log normal distribution, is laid on top of the graph. And it’s pretty accurate!
The research is similar to MIT Professor Devavrat Shah’s development of a new algorithm that can predict with 95% accuracy what topics will trend on Twitter an average of 90 minutes before Twitter’s own algorithm puts them on the list.
(Crystal ball image via Shutterstock.)