I’m no specialist in NLP but I imagine you could guess who is who by looking at the plots. To me, it is really funny how the density is very slim. Trump obviously goes for words that are far more emotionally charged either positively or negatively.
These sentiment scores come from a library called “bing”. It is the standard in the R package TidyText. There are other measures but I had no reason to not go with the default version.
The density might not be the preferred visualization so here is a histogram:
It is no secret the two presidential candidates have a very different way of communicating. One tweets very frequently. The other far less so. In this sample of 100 most recent tweets, the oldest Biden Tweet is from June 23 8 PM while the oldest Trump tweet is from July 4th 4 PM. That is a solid 6-day gap.
To visualise Traction I plotted likes and retweets for every send tweet from their official accounts. I know there are other people posting on these accounts but for my purpose here I don’t care about that. Since Twitter officially doesn’t allow for political adds to be pushed it is important the candidates earn attention. The more things are shared and retweeted the better.
If we look at raw numbers, Trump leads in Retweets, 2.2 million with Biden not that far behind 1.7 Million. Candidate Biden is leading in likes 8.2 million VS Trumps 6.7Million. In total Joe is winning in twitter engagement 9.7 million vs Trumps 8.8 million. Now it could be argued that timeframe matters. Someone might say that older tweets have a longer time to accumulate more likes and retweets. And that it would make sense to compare traction over the same timeframe. Which is a fair comment. However, not taking time into account and just looking at the most recent 100 tweets from both candidates have said is more interesting to me personally.
It does make it hard to compare both visually since Trump tweets multiple times a day.
What’s the Message?
What word is most likely to come from who?
I would say Joe talks a lot about Trump and Health and care (I imagine Healthcare in general) while Don is far more likely to talk about the media.
JOE’s Most Used Words
What stands out is that Joe talks a lot about Donald. Like, Really a lot. Perhaps more than Trump talks about himself. I don’t know how good or bad it is. Many of the most used words are related to the ongoing pandemic. For example, it is funny how the words “wear” and “mask” are among the most used whereas they are not among Trumps last 100 tweets. From these tweets, I can’t really make out what JOE’s big message is besides “Not Donald”
The last 100 tweets include the 4th of July weekend, as expected words like “nation”, “America”, “American” are used quite often.
Here is a word cloud(for words used >3 times) for the fans:
He talks a lot about people, America, The news and himself.
I like it how he does often use words that political consultants say work well with conservatives. “ Flag, God, Bless, Country, History”.
I was surprised that the word “Job”, “work”, “workers” and perhaps his favourite word “great” did not make the list.
Here is a wordcloud of Trump’s words (I am happy to see hydroxychloroquine is in there 🤣):
What Can We Learn?
Honestly not a lot of new things for those that are on twitter. If you are not on twitter this analysis saves you a lot of time and pain from reading all their tweets. 🤣
I Build a dashboard using R so you can access the analysis anytime. The next step is making it run in the cloud so that it does the analysis around 3 times a day. For the moment the word cloud is not rendering in the dashboard (it only does on my machine) I will eventually fix this. The dashboard does not include the log-odds part. I do not know if that would be a value add. Do let me know.
PS: I know it’s not the best looking dashboard on the internet, it’s only the second time I made one like this 😁😁