Voter Beware: The Media Slam Hillary Most....Says This Handy Computer Algorithm

April 24th, 2016 8:07 AM

Buyer beware: Not all studies of media bias and favoritism are equal. The liberal site Vox recently ran this headline. “Study: Hillary Clinton, not Donald Trump, gets the most negative media coverage.” Jeff Stein began:

The biggest news outlets have published more negative stories about Hillary Clinton than any other presidential candidate — including Donald Trump — since January 2015, according to a new analysis of hundreds of thousands of online stories published since last year.

Clinton has not only been hammered by the most negative coverage but the media also wrote the smallest proportion of positive stories about her, reports Crimson Hexagon, a social media software analytics company based out of Boston.

Stein wrote the data is based on hundreds of thousands of online news stories published since January 1, 2015. “1)The Huffington Post; 2) The Washington Post; 3) CNN; 4) The Washington Times; 5) Politico; 6) The New York Times; 7) Fox News; 8) MSNBC; 9) CBS News; 10) The New Yorker.

Wow, that sounds very definitive and thorough, except.....a computer did all the work:

Crimson Hexagon then took more than 170,000 posts by these outlets — stories published from January 1, 2015, until close to today — and ran them through their "auto-sentiment" tool. The software scans tens of thousands of stories within minutes for positive or negative language, sorts them into separate buckets, and tallies up the results.

For example, the software would take at a sentence that said "Trump made a stronger argument" and mark it as a "positive sentiment." Once it looks through the entire story, the software then categorizes the article as positive, negative, or neutral.

"We comb the content and see whether it's positive or negative," says Molly Moriarty, content marketing manager at Crimson Hexagon. "As you'll see, a lot of the conversation about the candidates is overtly negative."

....Crimson Hexagon created this software several years ago and it's mostly used to do market research for companies. To create it, staffers manually entered hundreds of thousands of stories and individually coded them as either "positive," "neutral," or "negative."

From this initial database, the company's software created patterns that now lets it automatically sort the posts by the "sentiment" of its content.

"Once you have done the human work to recognize and categorize all the posts, an algorithm can take over and — through pattern recognition — analyze an incredibly large dataset," says Benjamin Cockerell, director of global marketing at Crimson Hexagon.

Stein’s story never gave actual numbers, but a chart showed the differences between candidates was slight by this “auto-sentiment” measure. Just over 40 percent of Hillary’s coverage was negative, but about 35 percent of stories were judged negative for Trump, Ted Cruz, and Bernie Sanders. On the positive side, all these candidates seemed to get about four to six percent positive stories.

Liberals can’t really impress their friends from this number that Hillary is so much more badgered by reporters than the others.

Software isn't going to tell you that "Trump made a stronger argument" is a positive statement....if it was made by a white supremacist.

If the Media Research Center used its trusty method – actually reading the stories, one by one – and graded the tone as positive or negative, liberals could claim that an ideological bias would color that judgment, which is one reason why we don’t employ that method very often. Besides, you don't measure a liberal bias or a conservative bias solely by a positive/negative method.

As Stein acknowledges, the more realistic a candidate's chances, the more negativity they receive -- all the other candidates are gunning for them, and reporters feel the need to treat them more seriously. Hence, Sanders and Kasich measure more positively.

I am simply never going to trust a software algorithm that scans thousands of stories and judges them for you within minutes, any more than I would trust a software to tell me which Stephen King novel is more negative. Human perception is what clinches voting decisions.

You can see where this kind of software might impress a corporate client worrying about its public image, but to use it to measure media bias seems daft.