Newscasts

PNS Daily Newscast - November 14, 2018. 


Hate Crimes on the rise in the United States. Also on the Wednesday rundown: a big hearing in Denver on a proposed rollback of methane limits; plus find out about "Give to the Max Day."

Daily Newscasts

Forget Pollsters; Computer Science Uses Tweets to Size Up Voters

As Maine voters make their Election Day choices, computer scientists have created a program they say can take the nation's political pulse in real time using tweets. (Mike Clifford)
As Maine voters make their Election Day choices, computer scientists have created a program they say can take the nation's political pulse in real time using tweets. (Mike Clifford)
November 8, 2016

AUGUSTA, Maine – There has been no shortage of election polling as Maine voters cast their ballots today, but now there is a new way to measure voter attitudes by studying tweets.

Computer scientists from the University of Utah have developed what they call "sentiment analysis" software that can determine how someone feels based on what they write or say. In other words, Americans have learned a lot about how Donald Trump feels about people through Twitter, but how do Twitter users feel about Donald Trump?

Feifei Li, an associate professor at the University of Utah, helped develop the program, and said it provides a real-time window into how the public is reacting to political events.

"What's cool is that you can actually adjust the lens of the window," he said. "If you look at the last few months of data altogether, the sentiment for Democrats is stronger than the sentiment for Republicans."

Researchers said the biggest surge of positive tweets for Republicans came during their national convention, and again after the video was released of Trump boasting about sexually assaulting women. The peak for both positive and negative tweets for Democrats came on the heels of the final debates.

Li's team started with more than 250 million global tweets and used advanced software to identify 1.6 million political posts in the United States. Li noted the program was only able to detect the leanings of Twitter users, which are not reflective of the population at large. The group's results suggest that if larger data streams can be tapped in future versions, today's pollsters could find themselves out of a job.

"We feel like if you collect enough data from the population, if you get an unbiased sample from the population using this data-driven approach, the results might be more accurate," he added.

The scientists used artificial intelligence and machine learning to help the program decipher complexities in human language. The group then compared their results with the New York Times' forecasts and found the state-by-state analysis was very similar.

Eric Galatas/Mike Clifford, Public News Service - ME