For me, the interest combined with opportunity (work-related reason to pull out my sentiment mining code today). My findings here are interesting, though consistent with prior work (seriously, go read that prior post for context). In a way things got "worse." Here's what I found:
- Trump continues to be significantly more negative than the rest of the field.
- Trump continues to be significantly more bipolar in sentiment (higher sentiment variance) than the rest of the field.
- Trump is getting an even higher boost from his negative tweets now. Whereas historically he was receiving 230 additional retweets per negative word, he's currently getting over 800 incremental per negative word.
I'm using the same methodology as last time, downloading tweets using the Twitter API. Then I use fairly common stemming and word removal techniques, and throw my sentiment mining algorithm at it (code released in prior post). Also, this time I limited to July tweets, so I cut out all of the tweets used in my prior analysis.
The sentiment score derived is simply a score based on how negative or positive a tweet is using an algorithmic analysis of the words used in that tweet. More negatively scored tweets use more negative words, positive scored tweets use more positive words.
RESULTSThere are many ways to cut this data, but I will run the data few ways for quick analysis. Just quick data, findings and descriptions.
First, what do the distribution of tweet sentiments for each candidate look like? I left Bernie in because he's... still... apparently in the race. And myself, for full disclosure.
And a visual distribution of the two main candidates:
- Trump is the only one with negative average tweet sentiment.
- Trump also has the widest distribution of sentiment (visually, standard deviation), showing some bipolarity in his tweets.
- Clinton is the most positive tweeter of the group.
- I'm the most active on twitter of the group (hmmm).
Next we look at a more interesting area, how many retweets each candidate gets by net sentiment score. In essence, this analysis looks at how Twitter users react to each candidates tweet. The more retweets, the more interaction, the fewer.. less interaction. If we correlate this to the sentiment of tweets we can determine what kind of sentiment draws the most interaction for each candidate. Here's what Trump looks like (sentiment score is horizontal axis, retweets per tweet on vertical):
A few takeaways:
- The relationship for Trump is statistically significant, and we can infer that Trump gets 800 incremental retweets (on average) for each negative word he uses. (more negative == more interactions)
- The Clinton correlation is statistically insignificant, and almost completely flat.
And just for fun, let's look at the most negative tweet for each Tweeter:
Crooked Hillary Clinton is "guilty as hell" but the system is totally rigged and corrupt! Where are the 33,000 missing e-mails?— Donald J. Trump (@realDonaldTrump) July 4, 2016
"[@realDonaldTrump's] businesses were failing long before the rest of the town was struggling...his bad decisions hurt the whole city."— Hillary Clinton (@HillaryClinton) July 6, 2016
We have got to cut military spending around the globe and address the true causes of war: poverty, hatred, hopelessness and ignorance.— Bernie Sanders (@BernieSanders) July 5, 2016
Congrats to @DLind at @voxdotcom for writing the most intellectually condescending AND worst critique of Roland Fryer. Vox is horrible.— LB (@LeviABx) July 14, 2016
- Trump continues to be more negative on Twitter than Clinton, and also more bipolar. This behavior may be somewhat a reaction to his environment though, as Trump gets continuous positive feedback for worse tweets.
- Clinton's tweets continue to be more neutral, and somewhat less volatile in sentiment. Her followers also do not react systematically more positive towards more negative (or positive) tweets as Trump's followers do.
Oh, and I know you just hung around for the word clouds, so here you go.
Donald (Thanks people, Making America Great Again):
Hillary (mainly talks about Donald):