Analyzing Trump’s Tweets
A Data-Based Analysis of Trump’s Language on Twitter
President Donald Trump represents a lot of firsts. For my money, the most important of these, the one that will change politics for years to come, is his use of social media.
Never before has a sitting president been on Twitter¹. In generations past, it was difficult to figure out what the most powerful person in the world was thinking. We had to rely on press secretaries or political speeches. Now, these thoughts trickle out in a constant stream for all to see. A leader with a constant communication channel to the world is unprecedented. Even Roosevelt’s fireside chats were limited to about 3 a year.
With Donald Trump’s tweets, we have a LARGE corpus of text data² that can give us insights into the man’s mind. Below I analyze all of Trump’s (over 24,000) tweets. Let’s see what patterns emerge.
Reading Trump’s tweets, the first thing that jumps out is how simple his use of language is. If we run his tweets through the Hemingway app to analyze their complexity, we find that they average about a 5th-grade reading level.
Mind you, simple words are effective. Most marketing materials are written so that they can be understood by 6th graders³. If we look at the most common words in Trump’s tweets (removing stopwords like ‘the’ and ‘a’), we see that most are short and clear.
Polarization and Positivity
A lot of this simple language can be seen in Trump’s choice of adjectives. By far the most common word in Trump’s vocabulary is “great”. If we list the other common adjectives in his tweets, we see a study in simple opposites: ‘good’ and ‘bad’, ‘best’ and ‘worst’, ‘new’ and ‘old’, ‘fake’ and ‘fair’, ‘smart’ and ‘dumb’.
Just looking just at these adjectives, another trend emerges. Most of Trump’s descriptive words are positive (see those colored in blue). It seems that Trump’s boasts outnumber his attacks. Perhaps it’s not surprising that a student of Norman Vincent Peale’s “power of positive thinking”⁴ , Trump belives positive words can translate into positive outcomes.
If we use a naive sentiment classification to rate trump’s tweets, we again observe that more are positive than negative. This seems to be true even if we take negations (like “not good”) into account.
If we also plot this sentiment over time, we can map the history of Trump’s twitter presence. Below I show the average sentiment of Trump’s tweets for every month since his account opened in 2009.
We can see that early Trump tweets were largely positive and self-promotional. Trump’s earliest missives were quotes from his books or references to Miss America. Later, he would go negative in attacking then-president Barack Obama. Around the 2016 election, the sentiment of his tweets again went down. However, In the past few months, the valence of Trump’s tweets seems to be trending upwards. We’ll see if that continues after the November midterms.
Trump’s Enemies and the Subjects of Tweets
Though the average tone of Trump twitter is positive, his negative tweets are definitely the most memorable. In particular, Trump often turns to Twitter to attack his enemies. If we look at the most common people mentioned in his feed, we see that most are his political and social opponents.
Before he was elected, much of Trump’s output was directed towards attacks on President Obama (#1 and #2 on the list). Later, after he won the Republican primary, he turned his attention to Hilary Clinton (#3). Nevertheless, cultural icons are also targets of Trump’s ire. Interestingly, Bill Maher and Rosie O’Donnell get more mentions than even Robert Mueller.
Insult and Mockery
With so many political opponents, Trump has crafted a wide and varied array of insults to deploy against them. The most common insult words that he uses are shown below.
All of these attacks can be placed into 6 main buckets. These include accusations of weakness, stupidity, failure, illegitimacy, corruption or fear-based attacks. Some examples are below.
1 — Weakness— ie. low, old, lightweight, losing, losers, ridiculous, poor, pathetic
eg. “The U.S. has pathetically weak and ineffective Immigration Laws that the Democrats refuse to help us fix.”
2 — Stupidity — ie. dopey, incompetent, clueless, moron
eg. “Paul Begala, the dopey @CNN flunky and head of the Pro-Hillary Clinton Super PAC, has knowingly committed fraud in his first ad against me.”
3 — Failure — ie. failing, failed, disaster
eg. “No wonder the @nytimes is failing — who can believe what they write after the false, malicious & libelous story they did on me.”
4 — Illegitimacy⁵ — ie. fake, false, biased, hoax, haters, unfair, a joke, rigged
eg. “Wow, sleepy eyes @chucktodd is at it again. He is do [sic] totally biased.”
5 — Corrupt — ie crooked
e.g “Big story out that the FBI ignored tens of thousands of Crooked Hillary Emails, many of which are REALLY BAD”
6 — Fear — ie. enemy, threat, radical
eg. “Many of the Syrian rebels are radical jihadi Islamists who are murdering Christians”
Calls to Action
Analyzing the verbs that Trump uses most is as instructive as looking at his subjects and adjectives. These verbs are often used in calls to action.
Like any good leader (or demagogue depending on perspective), Trump needs to rally his supporters. Many of his tweets are devoted to this purpose. Besides the obvious “Make America Great Again”, here are some other examples:
“Congress must follow through on the Government’s plan on the Everglades Reservoir”
“A tragedy like Parkland can’t happen ever again!”
“Good Republican candidates will win BIG!”
“We will unite and we will win”
“I need your VOTE!”
Many of these calls to action are followed by words in ‘all-caps’ to add emphasis.
Mottos and N-grams
Besides finding the most common words in Trump’s tweets, we can also search for the most common phrases. In language processing, phrases of a specific length are called n-grams⁶. So for example, some interesting 4-word pairings in Trump’s tweets are shown below.
Besides ‘MAGA’, we also get “I will be interviewed” (usually on FoxandFriends), as well as “think like a champion” and “the art of the (deal)”, which are references to Trump’s books⁷.
The items in red represent Trump’s sayings, which he repeats often. These include, “the fake news media”, “america safe and great”, and “I will fix this”. We can see a similar pattern looking at 3-word phrases (trigrams) below.
Other Fun Facts
Twitter Use over Time
Looking at the dates of Trump’s tweets, we can see how much he has used the service over time. Believe it or not, Trump’s Twitter use has declined since he became president. It seems like it peaked in 2013 when he was A/B testing his Obama attacks and first dipping his toes in presidential politics.
Twitter Use Per Hour
We can also look at the times of day that Trump tweets the most. It seems that his Twitter use starts first thing in the morning and builds throughout the day, peaking around 3pm. A not-insignificant number of tweets are sent in the middle of the night too.
Others have attempted to map Trump’s social network by listing who he follows on Twitter⁵. I decided to go one step further by analyzing who he retweets. Below is a list of the most retweeted accounts in Trump’s Twitter feed.
Like most people, Trump seems to live in something of a filter bubble. His top retweets are from his own account and those of his aides, allies, or family.
Finally, since Twitter is an interest graph as much as a social one, we can get a sense of where Trump focusses his attention. So for example, we see that there are more mentions to ‘golf’ (314) than ‘border’ (296). There are also more references to ‘celebapprentice’ (260) and ‘apprenticenbc’ (152) than say ‘california’ (68).
A few years ago, most people barely knew what Twitter was. Today that same service is used by the most powerful person in the world many times a day. It contains all of Trump’s musings, jokes, opinions, attacks, propaganda and more. A sitting president has a bullhorn that he can use both as an outlet and as a means of driving the entire news cycle. Luckily all of this data is archived and easily parsed. I leave it to future historians to debate what it all means.
All code and data used for this project is on github at https://github.com/taubergm/TrumpTweets
1 — Yes @BarackObama has an account, but I doubt he ran it himself when he was president
2 — thanks to the Trump Twitter archive for collecting all of Donald Trump’s tweets in one place — http://trumptwitterarchive.com . They do their own analysis and it’s fun to look at the raw data
3 — Readability tests are defined here — https://en.wikipedia.org/wiki/Flesch%E2%80%93Kincaid_readability_tests
5 — For more examples of Trump’s attacks on the press, see this https://www.cjr.org/united_states_project/trump-twitter-spreadsheet-press-attacks.php
7 — Though the Art of the Deal is more famous, “Think Life a Champion” is one of many other books that Trump uses to frame himself as a success guru. See https://en.wikipedia.org/wiki/Bibliography_of_Donald_Trump
All book titles are concise calls to action.