By now you may have heard of IBMs supercomputer, Watson, from the game Jeopardy! and the ways it is being used to solve medical issues. But do you know Watson has gone into Twitter?(!)
Senior Editor at Forbes, Chris Helman recently shared how Watson is now running IBMs “Twitter Editor”.
When an oil company was getting crushed on Twitter about fracking they called on a team led by engineer Steve Welch, which seeks new ways to use the power of IBM’s Watson supercomputer.
The oil company hoped to use Watson to devise “the perfect tweet” to convince fracktivists” that the process was safe after over one million wells and fifty years.
But this story is not about fracking, it is about Watson and developing sweet tweets.
They thought they wanted to send the perfect tweet,” says Welch. In order for an oil company to connect with its enemies over Twitter, it needed to use the language that the community was using. “If you’re going to be on Twitter, you have to be authentic,” says Welch.
The goal was to have Watson “reverse-engineer authenticity”.
The project is five years old and Watson is very good at developing ‘golden’ tweets as well as genomics research and reading doctor’s notes.
“Watson is adept at reading fragmented language where traditional methods don’t work,” says Welch. The supercomputer teases out meaning and significance by looking at the context in which words are placed. “It turned out our technology was natural for Twitter because it didn’t have language structure either.” Like those doctors’ scrawled ciphers, tweets’ forced brevity ensures they contain words chosen with care.”
Helman tells us that today, Watson can link to a commercial feed of Twitter and see it in its entirety.
“The lesson, says Welch, “We’re already showing the limits of security by obfuscation.”
Back to assisting oil company and tweets about fracking.
“With that query, “what we were looking for is what gets people excited, what message is fueling the fire of the campaigns,” says Welch. To that end, IBM gave Watson a dictionary of keywords to start with, things like fracking, tarsands, poison, water, pollution, evil, shale, gas, fracked. Then they fed Watson 40 million retweets to ponder. They use retweets because an original tweet must have contained something resonant if it inspired someone else to retweet it.”
“Watson filtered out all the retweets that contained one of the keywords, then it catalogued all the other words appearing in those retweets. It logged how often those words are used and in what contexts and patterns. Based on its findings Watson assigns each word a “ringscore.” A word found to resonate more with re-tweeters earns a higher ringscore. “A tweet composed of high-ringscore words by definition is one that gets retweeted more widely than one with low-ringscore words.”
“With Watson’s help, the oil company can predict, even before it happens, which anti-fracking tweets are likely to go viral. And Watson can help the company craft its tweets, reflecting its own point of view, using high ringscore words most likely to resonate with the anti-fracking crowd.”
Helman shares that Watson now examines millions of tweets a day and that if your company has tens of thousands of dollars a month to spare, you can try it out yourself!
Welch shared a bit of Twitter wisdom we should all remember. According to Watson, there’s one low-ringscore term the use of which dooms a tweet to almost never getting retweeted: “LOL.”
Senior Editor Chris Helman is based in Houston, Texas. Follow him on Twitter @chrishelman.
Dr. Kevin Purcell, DC. Dedicated to serving others …