The Daily Northwestern

Northwestern faculty, students plan to improve ‘TweetCast’ ballot prediction app

Kristine Liao, Reporter

Hang on for a minute...we're trying to find some more stories you might like.


Email This Story






Solely based on tweets, an application developed by Northwestern professors and students is reported to predict people’s voting preferences with an 80 percent accuracy.

TweetCast is an online program that predicts who Twitter users will vote for by analyzing the terms, hashtags, websites and users mentioned in their tweets. It can also predict whether states will vote red or blue.

For example, terms such as “humanity,” “single” and “yall” are indicators of Hillary Clinton supporters, whereas terms such as “lying,” “corrupt” and “country” are indicators of Donald Trump supporters.

McCormick Prof. Larry Birnbaum and Shawn O’Banion (McCormick ’14) originally developed TweetCast for the 2012 presidential election. The program was also applied for 2013 elections in Norway and Germany.

“It was a cool project done for fun and a demonstration of what technology could do,” Birnbaum said. “We thought it was interesting and thought other people would think so too, and that turned out to be true.”

Thousands of people used the program in 2012, Birnbaum said. It was also featured on a PBS blog post titled “Our Picks for the Most Innovative Election Coverage,” and the 2013 Norwegian version gained attention locally, making headlines in Norway.

With the help of undergraduate McCormick students, Birnbaum updated the 2016 version of TweetCast to include a new geolocation feature, which makes it possible to predict whether states will vote red or blue. The state predictions, however, are not as accurate as the team would hope, Birnbaum said.

Jason Cohn, a fourth-year graduate student in the electrical engineering and computer science department, is attempting to improve the accuracy of TweetCast’s geolocation function.

“(The feature) is what I’d like to think (will) be the future of TweetCast, if we can do these mass predictions at the state level,” Cohn said. “It’s really an open question, and we’re not really sure if we’re going to be able to do it.”

One of the problems he faces is the demographic gap between Twitter users and the actual voting population. Twitter users are generally younger and more liberal, Cohn said.

To correct the demographic biases, Cohn is searching for variables that may reveal statistical relationships between the Twitter population and the general population.

McCormick Prof. Doug Downey, who is helping correct this demographic gap, said the application is a good way to predict people’s opinions without having to conduct costly polls.

“It’s actually very exciting, not just for this presidential election, but for estimating people’s preferences over a wide variety of issues in a much more cost-effective way than we currently do,” he said.

As long as a Twitter account is public, it can be subject to analysis, which Cohn said makes TweetCast an “accessible” application.

He said the project revealed to him how unaware people are about how much is available from their social media accounts.

“Just by reading 200 of your most recent tweets, we can predict who you’re voting for with 80 percent accuracy,” Cohn said. “We want people to think about this. There are a lot of implications, if you think about it.”

Email: kristineliao2020@u.northwestern.edu
Twitter: @kristine_liao

Comments