Rise of the Machines | Adweek
Advertisement

Rise of the Machines

Advertisement

In April, when Domino’s suddenly had to grapple with the fact that a YouTube video of a couple of employees doing disgusting things with the company’s food was circulating rapidly across the Web, it was bad for the pizza chain’s business.

But Domino’s problem turned out to be good for business for a fast-growing segment: companies that track Web chatter. In particular, text mining firms like Lexalytics, Clarabridge and Visible Technologies say they have seen a spike in interest for their automated software programs. Text mining, which is already used by some Wall Street traders to track issues that could affect stock prices, is now employed by many top marketers, including Cisco, Hormel, Microsoft and Intuit, as a sort of blunt instrument to gauge online sentiment about a brand.

While none of the companies claim their software can get a precise read 100 percent of the time, they do claim an accuracy rate of between 60 and 80 percent. Jeff Catlin, CEO of Lexalytics, which developed a proprietary text mining system for Cisco, said such a solution is needed because there’s just too much information—whether on blogs, in news stories, on Twitter or elsewhere—for human beings to digest in real time. “If you look at reading a single story, a human will be better than a machine,” he said, and then added, when it comes to reading 100 stories, “if a machine gets it right seven out of 10 times, you’ll be able to predict the overall sentiment."

Not everyone, however, believes that text mining is a silver bullet. Adam Nash, vp-search and platform products at LinkedIn, said his company has gotten a lot of requests from marketing and PR pros to add an interpretation feature to its popular Company Buzz tool, which tracks conversations on Twitter. But he thinks “it’s something that’s easy to do badly.” Steve Rappaport, knowledge solutions director at the Advertising Research Federation, meanwhile, cautions that text mining needs to be used with care. “It requires understanding what measures and indicators are truly measuring and indicating,” he said.

Continue to next page →