Romney ended the night as the undisputed champion of last night’s talking-point ping pong (er, “debate”). A post-event poll by CNN showed a Grand Canyon-size 2:1 gap among viewers in favor of Romney’s performance. But the big victory of the night may have gone to the University of Southern California for an experimental new social media monitoring tool, which showed a similar gap in the sentiment of tweets after the broadcast. Until now, attempts at deciphering public opinion from the young, liberally biased universe of social media have been dismal failures, but this new tool may be closer to cracking the problem.
Being able to divine public opinion from social media chatter could be incredibly helpful. Everyone from movie studios to the White House would instantaneously know how every slice of the American populace felt about every imaginable issue. Unfortunately, the cutting edge of statistical analysis just isn’t there yet.
“It can be concluded that the predictive power of Twitter regarding elections has been greatly exaggerated,” wrote computer science professor Daniel Gayo-Avello, after reviewing the current state of social media tea-leaf reading. Indeed, if social media reflected actual popularity, Ron Paul would be president, pot would be illegal, gay marriage would be a universal right, and Chick-fil-A would be bankrupt.
But last night, USC’s Twitter Sentiment Analysis tool, developed by their Signal Analysis and Interpretation Laboratory (SAIL) revealed a similar-sized gap in positive/negative tweets, as CNN found in its post-event poll. The technology is based on a strategy that matches its own analysis of Twitter opinion with the opinions of a real-life human. “Tweets in general, and political tweets in particular, tend to be quite sarcastic, presenting significant challenges for computer models,” writes the development team.
To get thousands of tweets evaluated cheaply by an actual person, they employed Amazon’s army of “Mechanical Turks,” known for quickly performing rote tasks, such as spell checking book reviews. The human-calibrated model is then deployed during real-world political events.
USC’s performance is promising, but it should be noted that other, more blunt measures of social media popularity also showed a big night for Romney. Topsy, a popular social media aggregator, showed a 2.5x spike in mentions for Romney over Obama.
So, while USC’s new tool should be taken with a large dose of caution, we could be much closer to realizing the Holy Grail of public opinion.
Oh, and for those commenters who want to complain about the accuracy of Romney or Obama’s arguments, or banter about who did better, there is only one true measure of “winning” a debate, and its who got the largest number of people to think that they won. This clip from Thank You for Smoking illustrates the concept beautifully, wherein a tobacco lobbyist teaches his son about swaying public opinion:
[Image Credit: Mother Jones]
[Via: The Hill]