Maybe you've wondered what Watson has been up to since he vanquished all comers on Jeopardy! a year ago. Primarily, as promised, the minds behind the supercomputer are working hard on medical software that will help doctors do deeper, more accurate and faster diagnoses. But a little bit of that power to compute in adaptive and detailed ways has also been applied to the Academy Awards this year, especially the social conversation around the nominees.
It's not that Watson needs to be entertained. IBM is developing a new tool that leverages social media buzz to help journalists tell a new kind of story and Hollywood to better connect with its audience. The algorithm can't predict Oscar outcomes, at least not yet, but it can help Hollywood understand earlier what will and won't work when its movies come out. And it can demonstrate, with far greater reach, how disconnected the Oscars are from the average movie-lover.
The idea for the research began as a news-gathering endeavor.
"We started with the concept of 21st Century tools for 21st Century journalism," says Steve Canepa, general manager for Media and Entertainment Industry at IBM. "The idea is that by using advanced analytics capabilities and our growing ability to analyze human language, we could improve the journalistic process and to distill insights out of all the data being created in the public."
The end product for now is the Oscar Senti-meter, a collaboration of IBM, University of Southern California Annenberg Innovation Lab and the Los Angeles Times. The Senti-meter has been analyzing millions of tweets since the nominations were announced, creating an aggregated voice of the people about the nominees. It turns out the Hollywood bigwigs who vote don't always see eye-t0-eye with the people who watch the movies.
"This project was a way to give this event a vox pop perspective," says Gabriel Kahn director of Future of Journalism at the Annenberg Innovation Lab. "The odd thing about this project and what makes it interesting, is the Academy Awards are decided by fewer than 6,000 people. So it's the ultimate elitist event, but it's followed by the world. We're looking at the 'Hurt Locker' disconnect here," referring to the Oscar winner that had not been widely distributed before it got the Academy lift.
For instance, one of the movies with the most positive sentiment is "Girl With the Dragon Tattoo," one at which Oscar shrugged his golden shoulder for a "Best Picture" nomination. Hollywood preferred Steven Spielberg's "War Horse." Twitter: meh.
"Spielberg may have been trying a little too hard," opined one tweep.
The Senti-meter discovered that there's not always a disconnect between the screeners and screened.
"Midnight in Paris" is a popular choice among viewers and Oscar voters, which gave the Woody Allen film four nominations, inlcuding "Best Picture." "The Help" also had Twitter and Oscar nodding in approval.
None of that is cause for much surprise, although a few items from the chart left me curious. Why did Gary Oldman get the worst sentiment of the whole batch? A look at the overall opinion about "Tinker Tailor Soldier Spy" might explain it, which mirrors the fact that Meryl Streep's tweep love is middling. Disliked movies seems to drag down the star's own sentiment ratings.
Journalism
But the real fascination comes from the algorithm behind the results and what it can eventually do for Hollywood and people who cover it. As the habit of posting thoughts about entertainment on social media platforms grows, the ability to capture and gauge that sentiment becomes more useful.
"If we could get a sense of the data around movies, what other opportunities does that open up to us?" says Kahn, who has worked as a journalist for two decades.
The labs is an interesting blend of programmers, analysts, writers and designers. And what should excite the news industry -- and its consumers -- about technologies such as these is the way a new kind of communication emerges.
"One thing we're all trying to improve here is audience engagement," Kahn says. "And I think being able to visualize something that is coming from the bottom up and giving your audience that data in a way that they can play with it, and manipulate and understand, is very valuable."
That data can surround almost any story.
"I'd really like to take this experience and point in different directions," Kahn says. "Another big story right now, for example, is the Republican primary and there's a different group (at Annenberg) looking at sentiment around the debates."
I can hear the rebuttals already. In fact, I recently read a debate among journalists on Facebook about whether quoting from Twitter was a legitimate practice. The anti-Twitter set was pretty sanctimonious, but Kahn doesn't buy it, as long as the journalist has the right tools.
"After seeing several thousand data points, you would have a very sophisticated understanding about whether quoting one tweet best represented a general sentiment," he says. "Journalists traditionally haven't hesitated gathering quotes from the street as a meaningful representation of broader sentiment, without these tools helping them determine whether it was accurate or not."
And the journalism can and would go beyond sentiment-gauging.
"There have been some really interesting studies about the kind of information that is flowing through Twitter," Kahn says. "One of the best examples is Osama bin Laden assassination."
In that case, Twitter users broke news throughout the night, including a unsuspecting live tweeting of the event. A highly sophisticated algorithm could quickly identify key sources of information, giving journalists the ability to filter out massive amounts of less important chatter. [NOTE: IBM is not currently working on this aspect of the news. It's purely a conjecture on my part about where the technology could go in the future.]
"But for the Oscars, which is the ultimate pop culture event, it doesn't work that way," Kahn says. "[The information] is too distributed. But there are also predictive possibilities for both journalists and the industry."
Hollywood
So perhaps an even more appealing outcome from this research is awaiting Hollywood. While, capturing sentiment is similar to diagnosing a patients' symptoms by cross-referencing them with massive datasets, using sentiment to predict outcomes could be a cure for Hollywood's ailing box office numbers. This is something Fizziology is already doing, using humans to analyze the data captured from Twitter and then predicting box office take from that aggregated sentiment.
The ability to predict is the result. For years, Hollywood had no empirical way to know whether their multi-million dollar ad campaigns would draw an audience until it was too late. But an analysis of millions of data points, using a tool both broad and fine in its sweep, is changing that prospect. It could also find the parts that are working and those that are not, helping Hollywood find and engage those most likely to buy a ticket.
"The result is seeing in real-time what their customers are saying and thinking," says Deepak Advani, vice president for Predictive Analytics at IBM, "and being able to react to that data by emphasizing the parts that customers reacted positively to and changing the things that they didn't."
Getting there is an never-ending set of adjustments. Consider the issues with accurately analyzing sentiment, using two well-known tweeps to tweeting about tonight's host, Billy Crystal.
When actor @LoganLerman tweets: "I'm so excited to see Billy Crystal host the Oscars tomorrow!" you can assume that is both a genuine sentiment and means something fairly important, because Lerman has 237,000+ followers.
But when comedian @robdelaney tweets: "OK! RT @BillyCrystal: @robdelaney Bro, I'm super hung over. Can you host the Oscars tonight? I'll owe you one. Thx." how does an algorithm pick up all of the factors that helps it not fall for the proverbial banana in the tailpipe? It's fictitious, it is meant entirely for humor and the reason you know that is because of its implausibility as a scenario and its overt use of unlikely language for Billy Crystal.
Apply this basic notion to a tweet about Hunger Games from @willpfeifer: "I was worried that, with Twilight coming to an end, I wouldn't have any incomprehensible pop culture to annoy me. So thank you Hunger Games!"
"This is the kind of thing [programmers] had to give a lot of attention to, for the Watson machine," Advani says. "But with IBM having several hundred mathematicians on staff, it allows us to crack these problems faster than most companies."
Canepa says sarcasm has been a major focus for the collaboration with the Annenberg Lab, and Advani says it's a matter of programming.
"You need to have people to train the system to understand what these things mean," Advani says.
Kahn is honest about the advancements thus far.
"The problem is comparing," Kahn says. "It's the old line about dipping your hand in the same stream twice -- it's not possible. It's too fluid and you don't have control over the response. But the upside far outweighs the downside."
And as Ken Jennings, and his cohorts of Jeopardy! champions, can tell you: don't underestimate Watson.
