Sentiment Analysis: Donald Trump & Hillary Clinton Tweets, Oct 5 – Oct 11, 2015

Emotion drives our decision-making. By appealing to emotion, others can persuade us to make decisions. We experience this during political campaigns.

Donald Trump knows the power of emotion. A charismatic leader, Trump infuses his speeches with appeals to emotion. Sentiment analysis makes this clear.

Comparing Donald Trump and Hillary Clinton, I sampled tweets from their respective profiles, published between Monday October 5, 2015, and Sunday October 11, 2015. Using sentiment analysis, each tweet was given a score between -1, the most negative, and +1, the most positive. Plotted across the 7 days, the results are displayed below, with Trump in red and Hillary in blue.
sentimentanalysisfinal

Trump exhibits a noisier sentiment artifact. Trump has almost no tweets with a sentiment score of 0. Trump peaks at +1 nine times; Hillary peaks at +1 three times. Using statistics, we see with Trump there is a greater range of sentiment, with a tendency towards positive sentiment. The Median Sentiment for Donald Trump is 0.21, whereas the Median Sentiment for Hillary Clinton is 0. The Standard Deviation for Donald Trump is 0.39, whereas the Standard Deviation for Hillary Clinton is 0.30.

Donald Trump Hillary Clinton
Median Sentiment 0.21 0.00
Standard Deviation 0.39 0.30

boxandwhiskers

So why does this matter? Noisy sentiment drives engagement. 

The chart below shows average tweet engagement for the respective profiles, for tweets published between Monday October 5, 2015, and Sunday October 11, 2015.

Donald Trump Hillary Clinton
Avg. Retweets 1028 783
Avg. Favorites 2136 1196

– Trump received 1.31 retweets for every 1 retweet Clinton received

– Trump received 1.79 favorites for every 1 favorite Clinton received

So while some political analysts doubt Trump’s ability to win over the Republican Establishment, these findings clearly show Trump resonates with the people who have direct access to him on Twitter. Like television before it, social media has ushered in a new era of political campaign strategy, and we must ask, how will this new means of communication influence the selection of the Republican Presidential Nominee.

Targeting an Audience, Mapping a Tour: Luther Dickinson

In this post, we will map Luther Dickinson’s US twitter followers, by count and influence, and examine how these distributions match his band’s upcoming tour routing, with the intent to demonstrate the value of twitter data for targeting audiences and planning performances.

Screen Shot 2015-08-11 at 10.37.43 PM

Luther Dickinson is the lead guitarist and vocalist for the North Mississippi Allstars. As of August 11, 2015, Luther has 1010 twitter followers. Of these 1010 followers, 483 identify their location as based in the US (not all followers identify location). The map below shows the concentrations of US followers, with the greatest numbers in darkest blue.

Followers

lutherdickinson follower map

Top 10 states by follower count (darkest blue):

State Followers
Tennessee 78
Mississippi 60
California 45
New York 35
Pennsylvania 25
Georgia 22
Colorado 20
Louisiana 18
Washington 17
Illinois 15


Now we will map Luther’s followers by influence, i.e. the followers of Luther’s followers. In other words, if each of Luther’s followers retweeted, how many individuals would see the retweet?

Influence

luterhdickinson_follower_influenceTop 10 states by influence (darkest blue):

State Influence
California 1137586
Tennessee 479783
New York 70690
Georgia 64776
Louisiana 63685
Mississippi 59011
Illinois 35373
Colorado 29206
Texas 26612
Rhode Island 23377

We see differences between followers and influence, with Mississippi, Pennsylvania, and Washington hosting greater concentrations of followers, who have less influence. Conversely, we see Rhode Island and Texas hosting lower concentrations of followers, who have more influence. California and Tennessee are strong points for both followers and influence.

Let’s see if this aligns with Luther’s plans for Fall 2015.

According to www.nmallstars.com, the band will tour the following cities in October 2015:

10.1 – San Francisco, CA
10.2 – San Francisco, CA
10.3 – Los Angeles, CA
10.4 – Anaheim, CA
10.5 – Solana Beach, CA
10.6 – Las Vegas, NV
10.9 – Boulder, CO
10.10 – Denver, CO
10.12 – Chicago, IL
10.13 – Pittsburgh, PA
10.14 – Washington D.C.
10.15. – Glenside, PA
10.16 – New York, NY
10.17 – Boston, MA
10.24 – Placerville, CA
10.25 – Placerville, CA

While we do not see a Tennessee performance during the stretch, all dates besides for two, in Las Vegas and Boston, match the list for top 10 states by followers. Furthermore, we see almost half of the performances, 44%, in California, a strong point for both followers and influence. We view this as strong support for the value of twitter data in targeting audiences and planning performances.

Luther’s map serves as a guide for up and coming artists within the genre. Using the raw data, one could target influencers within each state who would welcome the genre.

To this point, I envision developing a platform that leverages twitter data to help artists better identify audiences and geographic strong points within the genre. If you are an artist, manager, data scientist, or entrepreneur, and are interested in this work, contact me at andrewshamlet@gmail.com 

Part 3: Most common words used in tweets by Taylor Swift, Katy Perry, and Britney Spears

Hi there! For part 3, we will use visualization to analyze the most common words used in tweets among Taylor Swift, Katy Perry, and Britney Spears. In Part 1, we noted a favorite bias among entertainers. That is, entertainers receive more favorites than retweets. A clue to the favorite bias may lie in word choice. Let’s see.

Taylor Swift:

taylorswift_copyright

Katy Perry:

katyperrycopyright

Britney Spears:

britneyspearscopyright

Each visualization displays the most common words, by usage, for the respective twitter profiles during June 2015. The more often a word is used, the larger the word is displayed.

“Love” and “Day”/”Today” appear in large typeface on all three. Furthermore, the sense of temporality is evident among all three. We see words such as new, now, hour, tonight, and night.

Based on our findings, followers of Taylor Swift, Katy Perry, and Britney Spears receive strong messages of Love Now/Today/Tonight. There’s positive sentiment in these message, followed by an implicit call to action.

Once again, the charts from part 1.

Twitter10TenJune2015RetweetFavorite_v4

monthlyaverageretweettable_V2

Part 1: Top Twitter Profiles, June 2015

Hello! My name is Andrew Hamlet, and I am a MBA student at NYU. I am developing proficiency with Python, for data mining applications. I plan to publish research on this blog. Since my background is in social media and web analytics, that’s where I will start. As is the nature with scientific inquiry, collaboration is welcome! Please comment with suggestions or further lines of inquiry. Now let’s begin. Using Python, I gathered tweets, including the retweet and favorite counts, occurring in June 2015 for the Top Twitter Personalities (no companies), as listed by twittercounter.com. Here’s a table displaying the data, in ascending order by Followers. The Tweets, Retweets, and Favorites columns display totals for each Twitter Profile during June 2015. celebrityanalysistable_v2 Since there is variance among total number of Tweets, that is Katy Perry tweeted 35 times in June 2015 while Justin Bieber tweeted 167 times in June 2015, I normalized Retweets and Favorites, dividing them by Tweets to give Monthly Average Retweet per Tweet and Monthly Average Favorite per Tweet. Here’s a table displaying the results. monthlyaverageretweettable_V2 There appears to be a relationship Between Monthly Average Retweet per Tweet and Monthly Average Favorite per Tweet, that is as MARpT increases MAFpT increases. Let’s plot to verify. (For visual display, Justin Timberlake, Britney Spears, Ellen Degeneres, and Justin Bieber are removed from the chart.) Twitter10TenJune2015RetweetFavorite_v5 Yes! There’s a 0.83 correlation between MARpT and MAFpT, so Favorites increase as Retweets increase. Even more, we see a Favorite bias among this sample, that is the twitter profiles receive more MAFpT than MARpT. However, we see a Retweet Bias with Barack Obama. Perhaps politicians receive more retweets, while entertainers receive more favorites? In the next post, using this sample we will investigate whether there is a relationship between total number of Tweets and MARpT or MAFpT, that is does tweeting more during a month boost the number of interactions per tweet? Check back to find out!