When we were building SocialPatrol, our in-house software for social media management, artificial intelligence had yet to gain serious traction with companies like Facebook and Google. Fast-forward to 2017, and AI’s “Deep Learning” abilities have now exploded. The growth over the last five years has been tremendous; in 2016, a Google spokesperson stated that they had only two projects using “Deep Learning” in 2011, but now have over 1,000. And we’re using it too, as it’s the driving force behind our SocialPatrol.
Today’s AI and Deep Learning are positioned to revolutionize how we live and how technology helps us.
And, of course, our SocialPatrol is another example of this new “Deep Learning” era in artificial intelligence.
Think of SocialPatrol as eyes that never blink. If something’s said about our customers on any site we monitor for said clients, it’s seen by the all-seeing SocialPatrol. That’s when SocialPatrol’s algorithm thinks, “Whatchoo talkin’ ‘bout, Willis?” and analyzes that content, comment, or feedback, studies the context, then decides whether it’s positive, negative, or neutral.
Once SocialPatrol has classified the nature of that content, that’s where humans come in.
With pre-sorting done on comments, content, and peer-reviews for our clients around the world, our ICUC team is all over that content. They then scrutinize SocialPatrol’s sorting and decide what further action needs to be taken. Maybe that means escalating a concern, replying to feedback, or alerting the right people in our clients’ organizations. These actions can happen blazingly fast, often in as little as 20 minutes from when the public’s content pops up in, say, Boswell, Indiana, or in Dorset, U.K. Twenty minutes!
But here’s the really cool part...
Every single time that process happens – comment, sorted, actioned – SocialPatrol’s Deep Learning algorithm takes note of what the human operator did to close the case, and it helps to better inform the SocialPatrol sorting process the next time. It learns - and it learns fast.
SocialPatrol is multilingual and adaptable. So, whether it’s being employed by the Google Store to analyze user photos being submitted for customizable Nexus Live phone cases in less than 30 seconds from when the photo is received, or it’s scouring Disney fan sites for concerning content, SocialPatrol is a multi-level response system. We designed SocialPatrol to work with various pieces of software to gather such information - Crimson Hexagon, Sprout Social, native feeds from Facebook. You name it, we can work with it.
How this works is that we gather the initial content, but then SocialPatrol analyzes all of that for deeper meaning and relevance. So, instead of getting, say, 25,000 mentions of our client in a day from Crimson Hexagon, SocialPatrol’s job is to sort through all of that and decide the nature of those 25,000 mentions. Who’s happy? Who’s dissatisfied? What are the concerns being raised by our client’s customers?
Once SocialPatrol performs this analysis, it creates custom easy-to-read dashboards full of insights that will later allow for transparency and oversight at client level. But, first, the SocialPatrol AI team of veteran community managers scour the data compiled and sorted, and they train SocialPatrol’s AI to generate alerts based on client-specified guidelines. This is where the “Deep Learning” comes in.
Through their repeated input, our team of veteran moderators make SocialPatrol understand those client guidelines and teach it to decipher content for what’s really being said.
If you think about it, the English language can get pretty complicated, so SocialPatrol’s biggest hurdle is learning context.
So, here’s an example we love to use. Disney sees over 130 million visitors to their parks annually, so you can imagine their concerns about safety on their properties. Naturally, one of the terms we get SocialPatrol to search for is “fall.” SocialPatrol is able to scan a 10,000-word forum comment by a Disney fan in a nanosecond and decide that 20% of the content has potentially concerning feedback.
But then think about the words “fall” and “fell.” Context can drastically change their potential meaning. For instance, a fan can write, “It took me five seconds to fall in love with the park,” and it’s a wildly positive comment. To uninformed AI, the mere mention of “fall” could generate an alert and waste a Disney staffer’s time. That’s when our trusty team get on the case and educate the machine.
Now, through the human-aided sorting process, SocialPatrol learns that “fall in love” and “fell in love” are positive phrases, so is “fell for it.” “Fell hard” is a trickier comment that may need more instances of human clarification before the machine can discern between “I fell hard for the foot-long hot dog and snarfed down four in two days” versus “I stepped out of Magic Mountain and fell hard, gashing my knee wide open on an uneven step.” Obviously, we don’t need to send the litigation and maintenance teams an alert about foot-long hotdogs, thank goodness, but a head’s up on that Magic Mountain step could avert a huge lawsuit down the road.
Every time human moderators intercede on context questions, SocialPatrol remembers that action, and it informs future choices.
And that’s why SocialPatrol will always keep improving. When it comes to SocialPatrol, I think an old line of a Robbie Robertson song comes to mind, “You like it now, but you’ll learn to love it later,” because SocialPatrol keeps giving you more bang for your buck as it learns. As time goes on, our team can parse more and more information for you as SocialPatrol’s scouring and sorting gets ever more efficient.
Think of what that efficiency means to our clients’ bottom line: In 20 minutes or so, they’ll get a head’s up about problems or raves on their services or products. From averting lawsuits to, say, learning that there’s been a pricing glitch on an intercontinental flight that people are tweeting about, this can mean problems get solved in hours, not days. And, thanks to humans being involved in the review process, these won’t be false alerts – it’ll all be valid information that’s had the watchful eye of a human who understands, at an intimate level, what concerns our client. That multi-level process can be a game-changer for corporate response time.
Having a browser dashboard is great, but when feedback is flying from over 130 million visitors to parks a year, well, Disney Parks & Resorts need reports and information to be accessible on the fly, too. This is true for all our clients, large or small. As a result, SocialPatrol reports are available on desktop computers, in downloadable PDFs, through our easy-to-navigate smartphone and apps, and even in CSV reports for folks who love deep-dive analysis in minutiae.
When it comes down to it, we understand that our clients’ businesses live and die by popular opinion. Whether it’s happy customers or the first of a coming wave of irate feedback that may require a PR-level response, our clients need to know what’s getting said. Separating feedback only into “good, bad, and ugly” isn’t enough in this era of public sentiment writ large. Our clients need total knowledge about their customers’ experiences, because that affects everything from budgeting and marketing to staffing and beyond.
At ICUC, we believe in always being better. From SocialPatrol’s intelligence through to our clients’ performance, we’re committed to making “being better” a daily endeavor. SocialPatrol will never stop getting smarter, faster, or better… and our clients will, too.
In the next article in this series, I’ll talk about the public and business perception revolving around artificial intelligence and what it means for industries of all kinds. It’ll be a look at where the AI train is going, and both how and why you’ve got to get on board, no matter what industry you’re in.