Southern Cal Twitter-Mining Tool Picks an Oscar Dark Horse

Oscar statueWhen it comes to which movies deserve an Oscar this Sunday, are fans and the official voters divided?

The public would have picked Avatar over The Hurt Locker in 2010, some say. But this is “The Age of Big Data,” as The New York Times put it recently. Which means computers can help settle this film-buff debate, courtesy of a new tool co-developed by the University of Southern California’s Annenberg Innovation Lab.

The lab’s “Oscar Senti-meter” mines millions of Twitter messages to chart how much people are talking about the nominated films, and, more important, whether that chatter is positive or negative. The inside dope among industry types holds that The Artist will win best picture. And Twitter fans, too, are showing the silent film a lot of love, says Jonathan Taplin, director of the new USC lab.

“The dark horse, from our point of view, is Midnight in Paris, which also has a huge amount of positive sentiment,” Mr. Taplin says, repeating a prediction he made in Forbes on Tuesday. “That could be the one surprise.”

USC’s movie project, a partnership with IBM and the Los Angeles Times, grew out of a broader effort to apply the same technologies to the Arab Spring and politics.

The big challenge is sarcasm, Mr. Taplin says. When somebody tweets, “I’m so happy that Michele Bachmann threw her tin-foil hat in the ring,” the machine reads that as a sign of enthusiasm. But lots of human corrections have helped the computer get better at understanding the subtleties of language, as well as emoticons and other symbols.

With movies, the software has proven “astonishingly predictive” of success, Mr. Taplin tells Wired Campus. If a film comes out on a Friday, Twitter sentiment on the preceding Monday, Tuesday, and Wednesday can generally forecast its fate. For example, the studio thought last summer’s Cowboys & Aliens would be a smash. ”We knew from the sentiment that it was going to tank,” Mr. Taplin says.

His lab plans to release a tool similar to the Oscar Senti-meter for monitoring tweets about politics.

Return to Top