Some time ago, John Oliver did a pretty interesting and funny bit on his show Last Week Tonight about how science is covered in the daily press.
He points out how the importance or certainty of some results are overblown, and how often conflicting results arise.
The key thing when it comes to science is not so much the result of the first experiment, but whether or not the results of that first experiment are repeated every other time another person tries it. It’s similar to a key concept all of us learned in grade school math class: check your work.
One of the best parts of working on sentiment analysis is that it’s an emerging technology. What we can do is pretty amazing, and we’re only just scratching the surface. In fact, social media and how people consume content and engage with businesses may no longer be emerging, but it’s a rapidly evolving thing, too.
So, naturally, when we saw that Vox media had sentiment analysis done for the presidential primary we couldn’t help but note what a wonderful idea that was. Yet a number of folks I know (and probably folks you do to) were utterly disbelieving when they found that it was presumptive Democratic nominee Hillary Clinton was, by far, the most negatively covered candidate.
People have their preconceived notions about the press, and that it either has a bias one way or the other (liberal or for celebrity, in this case). So, it was easy to dismiss these results if they didn’t fit into their perception of how their favorite candidate was covered, whether it was Clinton or any of the other ones. Since sentiment analysis is an emerging technology, it can be easier to dismiss its relevance than one’s own political biases.
So, it was a nice bit of vindication for our industry when the second installment of the multi-part study of Presidential pre-primary coverage was released by Harvard’s Shorenstein Center on Media, Politics, and Public Policy. They found, like the Vox analysis found, that it was indeed Hillary Clinton who was the most negatively covered candidate of any party.
The study authors explain that this makes sense, because Hillary Clinton was seen, both in the polls and by the press, as the most inevitable winner on either side of the race. Because she has had a high public and political profile for the last 25 years, coverage of her is more critical. She was given the frontrunner’s scrutiny from the onset, and much of what she does is scandalized in the ideological press.
Ironically, it’s the very persistence of political bias that allows those whose own bias causes them to discount the previous results to discount this report as well. This study was conducted by people, the report written by people, and people have a political bias. Sometimes they aren’t even aware of it. A conservative who thinks his or her favorite candidate took lumps in the press more than Clinton would simply say this study was written by a bunch of liberal egg-heads in the tank for Hillary.
Yet, sentiment analysis software doesn’t have political bias. In fact, if the “liberal media” had that sort of ideological bias, the sentiment analysis would have shown that. It would have shown that the Democratic candidates get more favorable coverage than the beleaguered Republicans. But, instead, it showed the opposite.
Even though Vox and Harvard approached this same question in different ways, they were able to return similar results. This is pretty exciting because it helps us figure out how the press contributed to the election results. Only what’s even cooler is that, unlike in politics, the old guard and the new are able to help each other check their work.
Our math teachers would be proud.