Sentiment just before and after the election results, surprising absolutely no-one.

As soon as I saw the various news stories bubbling to the surface over at the Kaleida dashboard I knew I wanted to take a quick stab at the overall sentiment of news for the 24 hours before the results were announced/obvious and the 24 hours after, once the dust had settled.

Of course the usual broad brush of sentiment analysis should be take with a pinch of salt, see cravats below. To mitigate slightly the top row shows the sentiment for all news articles for a general baseline.

The “All news articles” are represented by a total of 961 articles gracing the front pages of the several news sources we track in the 24 hour period before the results, and 1,408 are from the 24 hours after. Which also represents a significantly higher than normal (and totally expected) churn rate for front page articles.

That’s our positive/neutral/negative baseline.

Now lets compare the overall positive sentiment with articles about Trump, Clinton and both. Before the results Trump was running at 75% negative 22.6% positive and hardly any neutral (2.4%!), Clinton was at just 11.1% positive sentiment. Which, putting my ‘Data Scientists’ hat on and not to get all technical, isn’t very much.

But after the results, boom, suddenly everything is a lot more positive, everyone loves Trump, everyone loves Clinton, everyone’s still not very keen on both of them together.

The best thing to do, based on this very limited information is to guess why all of this is the case in a way that aligns most closely with our own internal bias… and then take to twitter.

Or I could look deeper into the data and try and unpick which phrases and subjects are triggering the various sentiments, but it’s Friday, so that’s not going to happen.

Caveats

Here are the various disclaimers that need to be made.

1. Sentiment analysis is only as good as the source material, the corpus of data it’s been trained on, and how the computer happens to be feeling at the time. The analyser we use has been trained on news articles, as opposed to tweets, Facebook posts, Amazon reviews and such like. Those other sources have their own uses, but for us we really needed it to be based on news. With that said it’s still not perfect, and best used to spot over all trends rather than picking out and analysing specific pieces.
2. We scan the front pages of several major news sources from both sides of the political spectrum. So this isn’t all the news, just the news that made its way to the front page. For more details on the sources and sentiment see the previous article: A Jolly Jaunt Through the Sentiment of News.
3. In the 24 hours before the results we picked up a total of 961 articles from front pages, and 1,408 after. The number of stories that mentioned either Trump, Clinton or both before was 186 and 339 after. Which means election news was making up around 20% of the front page news before, and 25% after. Rather shockingly this means that a whole 75% of front page news wasn’t about the candidates, even if it didn’t feel like it.

Cravats