Dayton-area lab mines Twitter to make predictions

Credit: DaytonDailyNews

Cognovi Labs reads Internet tea leaves

Credit: DaytonDailyNews

Cognovi Labs’ tech predicts key elections

From data, leaders of local tech start-up Cognovi Labs say they read the future — confirming predictions about how people vote, what they buy and even whether they’re depressed.

Cognovi co-founders Amit Sheth and James Mainord are commercializing a technology Sheth and his Wright State University students developed, dubbed “Twitris,” to plumb the depths of social media for insights their customers can use.

The men say they predicted that the Republican Party would retain control of the U.S. Senate early on election day, Nov. 8, that Donald Trump would Florida, and that Trump would win the overall election, well before most in the media did.

Evaluating Twitter, hash tags, statements on Internet forums — and weighing the locations of those who posted all of that — Sheth predicted Trump would win Florida at 1 p.m. Nov. 8, hours before polls there closed.

At 9:22 p.m. election night, Sheth (@amit_p) tweeted: “Clinton’s path seems harder than Trumps to reach 272 #ElectionNight #Elections2016 : may be volume supporting him.”

Even before 5 p.m. on Election Day — with polls everywhere still open and most Clinton supporters still quite confident — Sheth saw uncertainty.

“#ElectionDay Twitter analysis of tweets of those #voted: #Trump has 75% volume, #Hillary has higher sentiment: confusing signal overall,” he wrote at 4:48 p.m. Nov. 8.

Between Oct. 13 and Nov. 8, Cognovi analyzed more than 60 million tweets.

Location matters in U.S. presidential elections. Cognovi saw early on that Clinton would win a simple popular vote, Mainord said.

But with the U.S. Electoral College, victories across a continent must be stitched together to win at least 270 electoral votes.

“It’s at the local level, the micro level, that we’re able to hone in,” Mainord said. “We were looking at people who said, ‘I voted.’”

Kim Frazier, director of growth initiatives at Dayton’s Entreprenuers Center, finds it interesting that social media sentiment is proving to be more telling than polls when it comes to predictions.

She also likes how Cognovi was able to harness where tweets originate in its insights.

“The way they shared the information, even down to the geographic area and the regions, state by state, was interesting also,” Frazier said.

The company also says Twitris read social media chatter well enough to predict that the “remain” camp was on its way to defeat in last June’s “Brexit” vote, the referendum on whether Great Britain was to retain European Union membership. Cognovi made the call hours before most news outlets did, the company says.

Working with Sheth, Mainord predicted the Brexit outcome hours before the outcome was announced in a LinkedIn post.

Brexit was Cognovi’s first real success.

Twitris’ potential extends beyond politics. Twitris can shed light on consumer reaction to product recalls and sentiment in other areas.

“We were really blown away by the capabilities it has and the opportunity to get a better understanding of the world around us,” Mainord said.

Mainord is managing director of IKove Venture Partners, which is licensing the Twitris technology from Wright State. He and Sheth started Cognovi in February, entering into an agreement with Wright State. The university owns the technology since it was developed at its Kno.e.sis Center.

“Things have changed,” Sheth said.

The ability to read the social media tea leaves is more useful than exit polling, Sheth argues. There are more tweets and statements on social media than poll responses.

And in the context of immediate political leanings, people tend to be more honest on Twitter, whereas some may lie to pollsters, Sheth also said.

“When they are volunteering themselves on social media, there is no reason for them to lie, at least in this context,” Sheth said. “There is fake news. There are a lot of lies. But at least in this context, there is a good chance almost all of them are real.”

The company is not yet in the black, Mainord said, but it has had a customer — he declined to identify — seeking election analysis.

Wright State, which licenses the technology to Cognovi, is a shareholder in the company, Sheth said. In return, the university gets some equity and “possibly” some percentage of revenue, he said.

Cognovi is structured so that the commercial beneficiaries are the entrepreneurs who developed the company, Sheth said.