The Signal and the Noise: Twitter, Elections and Word Clouds

A while ago, we built a site that looks the US Election conversation unfolding across Twitter. Here’s what social media tell us.

Digging through the data, Srinath Perera, our project head, extracted a list of the words most frequently used from Mid-March to the beginning of April.

Here’s a friendlier representation of this data.

Word clouds. Everyone likes word clouds. They highlight topics of importance. We can use this particular set of clouds to map some of the most important issues and incidents that happened during the election. Consider:

12th March, 2016. Chicago, trumprally. Thousands flocked to the University of Illinois in Chicago to protest a Donald Trump rally there.

17th March, 2016. Instagram. Russia. What do those two have in common? Simple: Donald Trump posted a bizarre video mashup of Putin doing judo, an Islamist militant gesturing with a gun and Hillary Clinton barking like a dog. Any Trump-Putin chumminess just went out of the window. Needless to say, the Kremlin didn’t like it.

22nd March. Brussels. Horrific. A one-word summary of the deadliest act of terrorism in the history of Belguim. Donald Trump used it as a justification of waterboarding. Ted Cruz hastily arranged a conference outside Fox News and stated that ‘Radical Islam is at war with us’ and used it to ask Obama to “immediately halt [his] ill-advised plan to bring in tens of thousands of Syrian Muslims refugees.” Hillary Clinton and Bernie Sanders were more toned-down in their messages; they spoke mostly of solidarity and loosely included a call to defeat terrorism as a global community.

26th March. Bird#rally. Bernieinseattle. Votetrump. In Portland, a bird flew onto the lectern Bernie Sanders was speaking from. He used this incident to good effect, playing it as a PR mascot. It turned into a meme of sorts — look up #BerdieSanders. The real kicker is that Sanders won three states, kicking up both victory cheers and a huge #votetrump cry from supporters on the other side of the political space.

That’s four instances where the word cloud’s correctly shown us what the world was thinking at that time.

Unfortunately, you’ll notice that there’s little insight to be gained from the rest of the sample space. Barring those four, the other clouds are tough to decipher. April 02. FBI. Is it because mentions of the FBI on Hillary’s case started sneaking back into news being written about her? Maybe. But it’s a maybe, not the more useful 1–1 mapping we pulled off earlier.

Consider the 23rd of March. Join. Join what? Join who? 25th. Path. Attack. Again, what? Who? Any correlation we can draw with this to events around the election are tenous at best. And this is the same for most days on the chart.

Does this make Word Clouds useless? Hardly. If you’re interested in what people are talking about in real-time, word clouds are one of the best options.

I think, however, that the cloud in itself is not a final piece of analysis. In our case, we can use word clouds to identify when our sample space start talking about major incidents — such as the Trump protest and Belgium. We can identify how interesting a given incident is to our sample space by seeing if it pops up in the word cloud. Everything else is a matter of watching the words and seeing what on earth they correlate to.

To look at the data for yourself, go to https://wso2.com/election2016/. Let me know what signals you see in the noise.