Modeling Twitter

We often hear that the Internet is unpredictable, that it’s the “Wild West.” That would seem to be especially true of a social medium such as Twitter. After all, tweets are by definition instant and short-lived. But in a paper I and my co-authors just submitted to the Annals of Applied Statistics, we describe a model we have developed that predicts how popular a tweet is likely to be within just a few minutes of when the “root tweet” is posted.

And anyone who wants to can now try out our model by visiting www.twouija.com.

Twitter usage would seem to be especially difficult to predict, but it turns out that there is systematic, repeatable behavior that can be modeled. My fellow researchers and I developed a dataset of 52 different tweets covering an array of topics, including music, politics, and miscellaneous everyday events. We then model how popular an original tweet will become based on the number of retweets within the first few minutes of its posting.

This research has potential implications for understanding the spread of broader ideas, memes, or trends in social networks and also revenue models for both individuals who ‘sell tweets’ and for those looking to monetize their reach. Knowing ahead of time how many retweets an original tweet is likely to receive makes it possible to determine which tweet can be used for purposes such as online advertisements and product recommendations.

I was surprised at how effectively we have been able to model Twitter behavior. It turns out that the Wild West is not so wild after all.

Tauhid Zaman is an Assistant Professor of Operations Management at the MIT Sloan School of Management. 

Tags: