The Cognoto team. From left: Brian Sacash, Joel Cornett and Sam Krassenstein.
Taylor Swift is stupid. Taylor Swift is stupid awesome.
According to startup company Cognoto, one of these phrases, while similar to the other, just doesn’t belong. (And hate TSwift all you want, but she’s not getting back together with you. Like, ever.)
Cognoto, which tied for third place during Baltimore’s Startup Weekend in September (with the original name FireTruck), would delete the first of the opening two sentences. Its overall goal is to reduce the number of negative comments typed underneath online articles.
It’s the brainchild of 26-year-old Brian Sacash, a Booz Allen Hamilton consultant and Ohio native now living just north of Washington, D.C. Sacash, whose background is in data analytics and quantitative analysis, was tired of checking online comments associated with articles only to find people insulting each other.
“Each word has its own meaning, but by itself, it’s kind of worthless,” says Sacash, who pitched the idea for Cognoto at Startup Weekend. “[We] wanted to find a way that we can look at word relationships and figure out how the actual sentence structure was being interpreted.”
That’s what Cognoto purports to do.
Essentially, it’s an algorithm that tries to examine a sentence typed online the way a person would interpret it. Just picking up on the word ‘stupid,’ for instance, is not enough for the algorithm to block a comment—it needs to see how ‘stupid’ is used in relation to other words.
A working beta product has yet to be built—it’s something Sacash says he and co-founders Joel Cornett and Sam Krassenstein are working on now—but the idea is to partner with large content providers and news organizations, who would then “ping” the Cognoto servers each time a comment is posted. Cognoto would then send back a percentage number for a particular comment, with a score closer to 100 percent indicating an OK comment.
However, what Cognoto will do begs the question: Who determines what is, and what is not, an acceptable comment? How do you differentiate between a blatant insult and, say, a political viewpoint that other commenters just don’t agree with?
“We would go to the content provider and say, ‘Where do you want to cut this off? Where is the line?’ ” Sacash says. “It depends on the content provider. You’re in their house, so you play by their rules.”
In other words, what’s acceptable for Reddit might be seen differently at the Baltimore Sun website. Either way, Sacash says the algorithm they’ve concocted can be “trained” to recognize acceptable and unacceptable comments.
“We give [the algorithm] examples of what is bad: this kind of sentence structure, these kinds of words. [We] go out and find examples of this,” says Sacash. “We also teach it what is not offensive. It gives it a reference point, [and] starts learning as you go forward.”
In the next month or so, Sacash and his team—who are all working part-time on the startup—hope to have a Cognoto website outfitted with a dialog box where people can test out the algorithm by typing in different comments and seeing which ones are deemed acceptable.
“Our goal was not necessarily to get rid of something that uses profanity,” he says. “We wanted to get rid of stuff that was not constructive.”