Predicting Retweets .png

Prediction of retweets of President Obama's tweet about his family dog

Some tweets get retweeted many times, some a few, and some just sit there. Now, researchers at M.I.T.’s Sloan School of Management have developed a model they say can predict the number of retweets a given tweet will engender -- and they can make the prediction within minutes of its posting. 

Sloan School Assistant Professor Tauhid Zaman, whose research focuses on social data analysis, said in a statement that many people “felt that Twitter usage was totally random and unpredictable, that it was all just noise.” But, he added, it turns out that there is “systematic, repeatable behavior that you can model” on Twitter. He has set up a site, called the Twouija in homage to the Ouija predicting board game, that graphs retweets over time as well as the accompanying prediction, for such Twitterers as Barack Obama, Will.i.am, Lawrence O’Donnell and Kim Kardashian.

The Not-So-Wild West

Based on a random sampling of the graphs, the predictions are not dead-on, but they can get pretty close. Actress Eva Longoria, for instance, was projected to have 295 retweets 179 minutes out, and actually received 271. At 178 minutes, President Obama was forecast for 478 and actually got 437. Many of the retweets build rapidly and then taper off.

It turns out, Zaman told news media, “that the Wild West is not so wild.”

The research team, consisting of Zaman and several colleagues, has recently submitted a paper to the Annals of Applied Statistics journal in which they detail their model. The implications of the research, Zamaan said, could impact our understanding of how memes and other trends spread via social networks, not to mention the effect it could have on the strategies of those who use Twitter for marketing.

Retweet Logic

To determine their model, the researchers collected data on retweets covering a large range of topics, such as music, politics and everyday events. The dataset is composed of 52 different tweets, and the model has two primary questions it asks -- when the “root tweet” was first posted and how quickly it was retweeted. How many retweets occur in the first few minutes, the researchers discovered, is the key factor in predicting how many retweets it will end up with.

With this kind of understanding, Zaman said, marketers can better predict new product adoption, political consultants can see the implications of a campaign theme, and demonstrators can understand if their cause has legs -- on Twitter, at least.