Tracking the Debate on Twitter: Southern Cal Tool Mines Political Sentiments in Real Time

Twenty-five thousands tweets per minute. That was roughly the volume of campaign-related Twitter messages posted during the height of the political conventions, according to researchers at the University of Southern California. Tonight’s presidential debate will unleash another Twitter tsunami.

So how do you make sense of it all?

A team of social scientists and computer scientists at the USC Annenberg Innovation Lab has developed technology that attempts to automatically track the sentiments of political tweets. The service, available here, displays the relative number of “positive” and “negative” tweets about each candidate, the overall volume of tweets pertaining to each campaign, and a scrolling list of the most positive and negative tweets.

The promise is that machines will mine Twitter to quickly detect changes in campaigns, before anyone has time to do polling and analysis, says François Bar, an associate professor of communication in the Annenberg School for Communication and Journalism.

But the project is a work in progress, he acknowledges, and one thing that causes “huge problems” is sarcasm. For example, during one of the primary debates, candidates were talking about invading Iran. Someone tweeted, “Oh yeah, we should invade Iran, it worked really well last time we did that in Afghanistan.” To the computer, that sounds positive, says Mr. Bar, who is working on the Twitter project in collaboration with the Signal Analysis and Interpretation Lab at USC’s Viterbi School of Engineering.

“The only way to know this is a sarcastic tweet is to really understand the context,” Mr. Bar says. “Which humans can do, but computers are really bad at doing.”

One way to deal with that problem: Have humans help train the computers. The developers have used real people to annotate a sample of some 15,000 or 20,000 tweets, noting whether each message was positive, negative, neutral, sarcastic, or funny. “We’re hoping, over time, to help the algorithm not be fooled by sarcasm,” Mr. Bar says.

You can help. Researchers are looking for volunteers to annotate tweets in real time during tonight’s debate. To participate, click here.

Return to Top