Social media's ocean of tweets and likes is supposed to be a treasure chest for companies wanting to target their online ads to consumers.
For example, if someone tweets how much they love a new movie trailer, it must mean that person is going to see the film in the theater, right?
Often, that's not the case. And companies are finding that parsing the meanings behind the global conversation taking place on social media isn't as straightforward as you might think.
A24, an independent film company in New York, is pitched constantly by sentiment-analysis services that promise to help companies better target ads on Facebook and Twitter, said co-founder Daniel Katz. But they rarely help lift ticket sales, he said.
"It's like opening your mail when you have someone who's promised you will win a million bucks," Katz said. "There's a lot of that. You're not actually getting a million bucks. It's an illusion."
- How Smart Meters Can Reveal Behavior at Home, What We Watch on TV
- 10 Surprising Ways Your Daily Life Is Feeding the Big Data Beast
- Cancer, Crime, Conflicts: Big Data Tackles World's Biggest Problems
Even social networks with the biggest stockpiles of personal data can have trouble pleasing advertisers. Darren Huston, chief executive officer of Priceline Group, one of the biggest spenders on online ads and a heavy investor in Google marketing, told Bloomberg News in April that his company has "endless amounts of money" to spend on Facebook and Twitter but they have failed to deliver results. "We haven't found anything there," he said.
Facebook and Twitter did not reply to requests for comment.
The trick to finding the hidden gems in social media is recognizing what to ignore. More than half of all posts are spam, auto-generated promotions, news headlines, reposts and other messages that are useless in determining consumer sentiment, said Ari Tuchman, a physicist who co-founded Quantifind, an analytics firm in Menlo Park, California. Nevertheless, much of that material is used in analytics services sold to companies, diluting the impact of their ads, he said.
Even posts that are gushing about products or companies often need to be discarded, Tuchman said. That's because paying customers signal their intentions in other ways.
For example: People who praise movie trailers hardly ever see the films, he said. But those who mention "babysitter," "girls' night out" or a relative's name do.
Quantifind also worked with a shampoo company that got a "massive amount" of positive buzz on social media about new scents they introduced - but no lift to sales, Tuchman said. The problem was that people enthusing about the product were excited about the scents in general, not the shampoo - and all the buzz was for naught. Those who bought the shampoo didn't mention the scents at all - they discussed its ability to repair hair damage, Tuchman said. His company maps clients' financial data to social-media mentions to identify conversations that led to higher sales.
People register their discontent in nuanced ways as well. Cell-phone users who blasted their providers about dropped calls seldom left those services. But those who use social media to complain about overage charges do leave, according to Tuchman.
"It's easy to find the things that you know you're looking for - it's much harder to find the insights from big data that aren't exactly what you know you're looking for," Tuchman said. "There are countless examples of more buzz not leading to more products being sold."
A24 hired Quantifind last year to help promote "The Bling Ring," and the software helped boost ticket sales and reduce promotional costs, Katz said.
Separating meaningful signals from the noise will be especially important in overseas markets. For example, most of Twitter's users are outside the U.S., but only a quarter of the company's revenue is generated internationally.
But no matter the language or the medium, the evolving science of sentiment analysis shows that even amid the volumes of data being analyzed, people are still hard to read.