Predicting the Popularity of a Tweet

Who’s more influential among their Twitter followers, Barack Obama or Kim Kardashian? You might not want to find out.

While Obama outnumbers Kardashian in total followers, 32.6 million to 17.9 million, research by MIT alumni suggests that Kardashian’s tweets may have a much stronger effect on her social media audience.

A group of researchers that includes MIT Assistant Professor Tauhid Zaman ’04, MEng ’05, PhD ’11 and University of Washington Assistant Professor Emily B. Fox ’04, MEng ’05, EE ’08, PhD ’09 have created a prediction tool that can estimate a person’s rate of being retweeted.

The tool—which the team calls “The Twouija” (rhymes with Ouija)—predicted that an April 15, 2012 Kardashian tweet would generate about 800 retweets in about an hour while an Obama tweet from the same day would generate about 440.

The Twouija predicts the popularity of a tweet shortly after posting. According to the study, most Twitter messages have the same lifespan of newsworthy-ness and a tweet’s popularity can be predicted within the first five minutes.

From “A Bayesian Approach for Predicting the Popularity of Tweets:”

We measure the popularity of a tweet by the time-series path of its retweets, which is when people forward the tweet to others. We develop a probabilistic model for the evolution of the retweets using a Bayesian approach, and form predictions using only observations on the retweet times and the local network or 'graph' structure of the retweeters.”

The group has showcased their research at a website they call The Twouija: Retweet Oracle, which graphs the timespan of retweets (actual and predicted) from the Twitter messages of 40 well-known personalities such as Seth Meyers, Diddy, and Newt Gingrich. (Hey—where is @mit_alumni?)

Retweets predictions for musicians Will.I.Am and Pitbull were effective, but much like its Ouija namesake, the oracle is not always precise. In one model, it estimated that a tweet from Eva Longoria would be retweeted 254 times within 60 minutes. In reality, it was only retweeted 183 times.While retweet prediction is scientific, it's also subjective. Much depends on the content of the original tweet.

As Zaman notes in the report, an efficient retweet predictor has serious implications for understanding more about internet virality and how social media revenue models can be better monetized. (In June 2012, Kardashian was paid $10,000 for a tweet about shoedazzle.com.)

Despite Kardashian’s online popularity, she did not have the highest prediction of retweets in the Twouija’s sample. That honors goes to The Rock, whose April 15, 2012 tweet, “Good morning! Enjoy your Sunday. #Faith,” had more than 850 retweets within the first sixty minutes. (The Twouija predicted about 1,000.)

Read the heavily scientific 28-page paper, "A Bayesian Approach for Predicting the Popularity of Tweets," which also includes research from University of Pennsylvania Professor Eric T. Bradlow, at the Cornell University online library.