Tweeting the elections

Northeastern researchers created a “wordcloud,” in which the font sizes and colors of keywords reflect the relative popularity of words mentioned at least once on a Web site for a political candidate. Courtesy photo.

Do tweeters in Texas care about taxes more than tweeters in Tennessee?

Over the last two days, a team of researchers from Northeastern University, Harvard Medical School and the Technical University of Denmark tracked the language of more than 20 million tweets to capture the political mood of the Twitterverse before today’s midterm elections.

“It’s well known that the state you live in plays a role in deciding what issues you care about,” said Alan Mislove, an assistant professor of computer and information science at Northeastern University who contributed to the study. “Because conversations on Twitter are public, we can geocode individual tweets and study where Americans are talking about specific issues.”

Earlier this year, the team analyzed the language of 300 million tweets to measure the collective happiness of people across the country. They found that tweeters on the West Coast are considerably happier than those on the East Coast.

For this project, researchers built a system to scan tweets in real-time for politically charged keywords, such as “unemployment,” “climate” and “terrorism.” They chose words by mining about 2,140 websites for all 1,152 candidates for governor and Congress.

Researchers geographically represented the data using a density-equalizing map, in which each region is scaled to represent its number of tweets as opposed to its land area.

Twitter Nation appeared most interested in job- and gay rights-related issues, and least interested in abortion- and veterans-related issues, according to a crude visual analysis of the map on Monday afternoon.

But, as Mislove put it, “We can’t say that we see any particular trends just yet.”

In another study, the same group of researchers found a large gap between the most popular political topics among Democratic and Republican candidates, according to a daily word analysis of their websites.

Researchers said Democrats expended more text on education and jobs, while Republicans devoted more attention to taxes and immigration. Neither Democratic nor Republican candidates talked about the wars in Afghanistan or Iraq or gay marriage, they said.

Researchers mapped their findings using a so-called “campaign wordcloud,” in which the font sizes and colors of keywords such as “house,” “Washington” and “country” reflect the relative popularity of words mentioned at least once on a website.

“Digital traces are a powerful, yet simple approach to providing news to a mass audience in the same way that public opinion polls are useful for gauging voters’ interest in a particular candidate,” said David Lazer, an associate professor of political science and computer science at Northeastern University who contributed to the study. “These interactive tools are quite useful for consumers of news who want to understand the dynamics of elections in both temporal and geographic ways.”

Other collaborators on both projects include Yong-Yeol Ahn and Yu-Ru Lin, postdoctoral research associates at Northeastern’s Center for Complex Network Research; Jukka-Pekka Onnela and J. Niels Rosenquist, both of Harvard Medical School; and Sune Lehman, an assistant professor of informatics and mathematical modeling, at the Technical University of Denmark.

View selected publications of David Lazer in IRis, Northeastern’s digital archive.