You are here: Foswiki>Dmi Web>WinterSchool201710minutespostelection (20 Feb 2017, NataliaSanchez)Edit Attach

“Trump didn't win. Racism won. Sexism won. Hate won. Lack of education won”

Paola Villarreal on Twitter- -Not  A. Single. Person. Of. Color... S. election- The world reacts to Donald Trump's surge - Nov. 9, 2016.jpg
Team Members

Natalia Sanchez, Humberto Ferreira da Silva Junior, Stefania Guerra, Willem van Hees, Dağhan Irak, Tatiana Terra Ruediger, Marit Willemsen, and Emma van der Zalm

Summary of Key Findings

The most important findings is that after the Wisconsin victory, it came clear that Trump would be the new US president. An exploratory and triangulated research is used in order to reconstruct the election night through sharing emotions on Twitter through tweets, words, and links. A URL-study showed that first people mostly shared live blogs of media outlets, like CNN and Washington Post. There is a time-period transition (08u00-08u30) where the Americans go to sleep and the rest of the world wakes up, picking up shared links. Besides the frequently shared news announcements, first emotional reactions on twitter are curious regarding the future, followed by primal extreme feelings. Negative feelings are described as desperation and subcategorized as ‘angry-disbelieve’, ‘disgust’, and ‘rebellious revolting’.

Table of content

1. Introduction p. 3

2. Initial Data Sets p. 4

3. Research Questions p. 4

4. Methodology p. 5

4.1 Methodology Analysis of the first fifteen minutes p. 5

4.2 Methodology Discourse Analysis of tweets p. 6

4.3 Methodology URL-Study p. 6

5. Findings p. 7

5.1 Findings Analysis of the first fifteen minutes p. 7

5.2 Timeline emotions p. 9

5.3 Findings Discourse Analysis of tweets p.10

5.4 Similar result Sentiment Analysis and Discourse Analysis p.13

5.5 Findings URL- Study p.14

6. Discussion p.18

7. Conclusions p.19

7.1 First fifteen minutes p.19
7.2 URL-study p.19

7.3 General conclusion p.20

8. References p.20

Appendix A p.23

Appendix B p.24

Appendix C p.25

1. Introduction

U.S. elections have been key subject of (social scientific) research (e.g. Agenda-setting theory by McCombs and Shaw, 1972). With the new eras of the Internet and Social Media,

it has been well-documented that Twitter has become a staple of political debate, notably before elections (Vergeer and Hermans, 2013; Christensen, 2013; Bekafigo and McBride 2013). However the role of Twitter in such political events is usually confined to “predicting elections,” which, as Gayo-Avello (2012) states, is a fruitless effort. On the other hand, it is also documented that Twitter may act as a great tool of social research during or in the aftermath of important events, such as earthquakes and other natural disasters (Earle, Bowden and Guy, 2012; Sakaki, Okazaki and Matsuo 2010; Crooks et al. 2013). Nevertheless, the reactions of users during political events belong to rather uncharted territories. The aim of this subproject is to examine the reactions of Twitter users in the US Election night, in two levels. First, the reactions during the first fifteen minutes after Donald Trump was declared to be the president-elect (2.30AM – 2.45AM Eastern Time) are examined through combining qualitative and quantitative analysis methods. As a secondary study, the evolution of the URLs shared through the night (10PM – 6AM Eastern Time) is examined to detect any changes in the content shared in different periods of time.
Messages on Twitter, so-called tweets, can be seen as mini-discourses (Bouvier, 2015; Chiluwa, 2015) and Twitter provides a platform to post information by the public, media and organizations (Van der Meer, in press). Discourses portray values, norms, judgements, and ideologies through lexical items (Van Dijk, 1995). Discourse analysis researches qualitatively “language-in-use” (Gee, 2011) and maps the interplay of hegemonic discourses (Catenaccaio et al., 2011). Therefore, discourse analysis offers a second and different perspective of the tones of voices of the tweets, to portray their emotional spheres from another angle.

2. Initial Data Sets

For the study, a dataset, comprised of tweets collected by the DMI-TCAT tool on the US election night (8-9 November), based on keywords was used. While the dataset contains tweets sent between 10PM and 6AM (ET), only the URL study used the entire database, while the ten-minute study used a filtered version of the database that features tweets sent between 2.30AM and 2.45AM. The dataset contains information on tweet content, date-time information, user information, and sentiment analysis, and discourse analysis.

3. Research Questions

The general research question is:

How can the U.S. election night 2016 be reconstructed, based on the transition from sharing facts (such as amount of votes) into sharing emotions on Twitter through tweets, words, and links?

To answer the general research question, we formulated the following sub research question:

What do the reactions about the U.S. elections on Twitter in the first fifteen minutes after the victory look like?

4. Methodology

Since we discussed that one single method isn’t appropriate and sufficient to reconstruct the election night, we used the basic principle of triangulation by using different research methods in order to come to the findings. Thus, we created the idea that the subject of research could be portrayed in three layers:

4.1 Methodology Analysis of the first fifteen minutes

(1A) A sentiment analysis

(1B) Timeline emotions

(1A) and (1B) are examined by a digital tool based on the NRC Word Association Lexicon (EmoLex) in TCAT. In order to reach the results, firstly, we’ve determined the exact time Trump passed the threshold of 270 electoral votes. That happened when he defeated Hillary Clinton in Wisconsin at 2.30 am, Eastern Standard time. Next, we collected a sample of Tweets using DMI Twitter Capturing and Analysis Toolset (DMI-TCAT). The selection of tweets about the U.S elections was made by using the keywords ‘Clinton’, ‘Hillary’ and ‘Trump’.

After the data was collected, we made a word cloud (see appendix A) containing the most frequent words used in the Tweets send during the 10 minutes after the election-outcome. The word cloud distinguished clearly two themes: “sexism” and “racism”. The sexism theme contained words like, “man”, “woman”, “sexist”, and, “sexism”. The racist theme contained words such as “race”, “racism”, “black”, and, “white”.

With TCAT two new samples were selected, based on the frequency of retweeted tweets. Leading to a corpus of 2895 tweets. The argument here is that when a tweet is retweeted frequently, it channels a point of view or emotion. We read, and selected theme-oriented tweets, and deleted tweets which didn’t correspond with any theme. in this way 54 tweets regarding “sexism" (n = 54) and 79 tweets regarding “racism” (n = 79) remained.

In order to analyse the categorized tweets on both themes, sexism and racism We used a sentiment analysis tool to analyse those tweets. The tool is based on the NRC Word Association Lexicon (EmoLex). According to the EmoLex website: “The NRC Emotion Lexicon is a list of English words and their associations with eight basic emotions (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust) and two sentiments (negative and positive).”

4.2 Methodology Discourse Analysis of tweets

Based on an a-select sample of the tweets (n = 2856) a discourse analysis was made.The tweets were imported in Microsoft Excel and sorted ascending on alphabet. In this way (1) all the most frequented and same retweeted tweets were clustered, (2) the most frequent retweets were clustered, and (3) other salient tweets were more easy to be reviewed.

4.3 Methodology URL Study

For the URL- Study, first, we divided the data set in half hours, which resulted in fourteen datasets. Then, we filtered only the columns including only the text of the tweets and the time there were posted. Using the DMI Tag Cloud Tool, we extracted a word cloud of each half hour. Then, a qualitative filtering was made, excluding general words, such as the name of the candidates and general hashtags. Using this new spreadsheet with clean and filtered data, we developed a visualization overtime with the new data set to have a better a understanding of the election night. method. We first extracted the URL frequency numbers per interval from TCAT, compared the totals and created a “URL frequency over time” graph, resulting in Figure 3. that we would take as starting point for the URL-study.

---++ 5. Findings

5. 1 Findings Analysis of the first fifteen minutes
5.1.1 Sentiment Analysis of the tweets

Overall sentiments of all the tweets sent in the 10 minutes after it became known that Trump had won the elections were divided in positive and negative. Our findings show that for all those tweets, the prevailing sentiment is positive rather than negative. However, looking at the tweets about racism and sexism, the sentiments differ slightly. The sentiments of those tweets are still more positive than negative, although there is less discrepancy between the two.

For the top 10 most retweeted tweets, the overall sentiments were positive. The tweets show mostly news announcements and also some tweets congratulating Trump with his win. On the other hand, the sentiments of the top 10 most retweeted tweets about racism and sexism are mostly negative. This can be explained by the fact that the racism and sexism themed tweets contain more emotions overall, they are less factual than the top ten retweets overall. Looking at those tweets, more or less all of them are negative towards Trumps win and accuse Trump of being racist and sexist. Also they put forwards their worries on the subjects.

As with the top 10 most retweeted tweets, the overall sentiments of the top 10 tweets with the most followers, were positive. This was due to the tweets being mostly factual and coming from news organizations. Looking at racism and sexism themed tweets, they were more positive compared to the top 10 most retweeted tweets. This is probably because the tweets contain a more nuanced tone. However, the amount of tweets with a more negative sentiment is not significantly lower than the amount of tweets with a more positive sentiment.

5.1.2. List of emotions and general tweets

All tweets show mostly trust, indicating a lot of news containing tweets. Other, less frequent, emotions are “anticipation” and “sadness”. For the sexism and racism tweets, the similar emotions prevail: “joy”, “anger”, and “trust”. The categorization of these emotions is almost similar. Salient notion is that opposing and extreme emotions “joy” and “anger” are represented, indicating that the themes racism and sexism evoke extreme emotions.

5.1.3 Top ten retweets

Again, for all tweets, the main emotion is trust, again indicating a lot of tweets with news content. Also, anticipation and joy prevail. This is illustrated in the following tweets:

(1) “America, meet your new president: Donald J. Trump.

(2) “Imaginez Trump , Poutine et Marine Le Pen mdr c'est des bz ils vont jouer au foot le samedi avec la bombe nucléaire”.

However, in case of all the tweets, the tweets contain much less emotions than the tweets about racism and sexism. That can be explained by the fact that the overall tweets are mostly factual and racism and sexism are more emotionally loaded topics.

The following salient tweet was retweeted the most: “Trump didn't win tonight. Racism won. Homophobia won. Islamophobia won. White supremacy won. Sexism won. Sexual assault won. Hatred won.” This tweet contains a lot of different emotions, adding on the overall level of emotions in racism and sexism themed tweets.

5.1.4 Top ten followers

For the top ten followers tweets, anger, fear and trust prevail as emotions in the overall tweets. Those tweets are linked to news agencies tweeting about the news of Trumps victory. They mostly contain information on Clinton conceding the race, which might explain the emotions of anger and fear.

For the tweets on racism and sexism, there are spikes in the emotions of joy, anticipation and trust. This is striking because looking at the tweets, they are quite negative about Trump’s victory. However, an explanation might be that those tweets are more nuanced, and less emotional overall than the top 10 most retweeded.

5.1.5 Most retweeted

The most retweeted tweets show almost the same division of emotions, except for the tweets on sexism.

5.2 Timeline emotions

When comparing the timelines of emotions of all the tweets sent within the first fifteen minutes after Trumps win and the tweets about racism and sexism we saw the following differences.

For all tweets, the first minutes show a big peak in tweets with the emotion trust. One explanation might be that most of the tweets sent during this particular time slot, use words that according to the EmoLex, are associated with trust. As seen in the EmoLex, words like ‘win’ and ‘winning’ are associated with trust. The majority of tweets during the first couple of minutes after Trumps win, might have had to do with ‘winning’.

Another explanation might be the number of factual newstweets after the election result. When you look at the graphs that show the overview of emotions during those 15 minutes, the emotion ‘trust’ spiked when the tweets were more or less factual newstweets, sent by news organizations. After two minutes, people could have switched from posting tweets that contained the news of Trumps victory, to tweets that are more emotional. This is shown in the graph as a point at which the ‘trust-tweets’ fall and the other emotions are divided proportionally amongst all tweets sent. We have to note, however, that these explanations are based on our own interpretations of the findings.

For the tweets about racism, there are a few peaks in emotional tweets, but only after about 13 minutes, there is a big peak in tweets containing more emotions. Except for in the first two minutes (during which joy and trust prevail) all emotions are divided quite proportionally.

For the tweets about sexism, they show less peaks. However, same as the tweets about racism, they show a big peak in emotions after approximately 13 minutes.

5.3 Findings Discourse Analysis of tweets

The tweets were categorized on the following content:

(1) Media news outlet (33.58%) such as: “Fox News: “Donald Trump wins the White House jxonMbsoY0 | #NewsInVids”;

(2) Pro-Trump tweets (23.04%) such as: "Trump gets past the polls... OOOHHH YEEEEES. WELCOME TO THE WHITE HOUSE DONALD TRUMP.”

(3) Support Hillary-tweets (9.67%) such as: “RT @ACAttaway: Hillary is proof a woman can work hard, rise to the top of her field & still have to compete against a less qualified man for the same job.”

(4) Anti-Trump-tweets (33.71%) such as: “RT @danacfinley: If you want to understand why rape culture is alive in the USA, it's because Trump can admit to sexual assault & still get millions of votes.”

Figure 1. displays these main findings of the four categories in a graph.Willem01.jpg

Figure 1. Main findings of the four categories

Due to limited time, we choose to focus on one of these categories: the anti-Trump-tweets. The following three emotional spheres are based on based on highest frequency retweet:

  • (1) Angry Disbelieve (11.38%) such as: “RT @Y2SHAF: If Donald Trump wins the presidency with no background in politics I shouldn't have to graduate university to get a good job in my field.”
  • (2) Disgust (9.1%) such as: RT @Moouxe: #Trump didn't win. Racism won. Sexism won. Hate won. Lack of education won. #ElectionNight #ElectionNight
  • (3) Rebellious Revolting (8.6%) such as: “RT @apblake: "Fuck Donald Trump" chant outside White House right now
  • The next emotional spheres are selected on salience:
  • (4) Absurd (2.35%) such as: “RT @grilcode: IF DONALD TRUMP WINS ILL SEND EVERYONE $100 WHO RT'D THIS.. NO JOKE”
  • (5) Fear (2.28%) such as the next tweet: “RT @farwzaz: "Trump's win means we'll fight harder tomorrow. I'm tired of fighting. I'm scared as shit. it's not my job to fix this.”

Figure 2. illustrates these Anti-Trump tweet findings in a graph.


Figure 2. The anti-Trump-tweets.

5.4 Similar result Sentiment Analysis and Discourse Analysis

As well the sentiment analysis as the discourse analysis point the next salient, high frequent tweet or variations of it:

“Trump didn't win. Racism won. Sexism won. Hate won. Lack of education won.”

The sub research question: “What do the reactions about the U.S. elections on Twitter in the first 10 minutes after the victory look like?” can be answered as followed:

There is a wide range of extreme emotions and sentiments, both negative as positive, such as “joy”, “anger”, and “trust” . The emotions represent mostly primal feelings and thoughts, that can best be described as “desperate” like angry-disbelieve, disgust, and rebellious revolting. This is best illustrated by the highly retweeted tweet: “Trump didn't win. Racism won. Sexism won. Hate won. Lack of education won.”

5.5 Findings The URL - Study

The goal is to detect the frequencies of URLs shared on Twitter, since the beginning of the election night (3AM UTC) to the end (11AM UTC) in 30-minute time intervals. Figure 3. illustrates that there are two rather silent periods; between 06.00-06.30 and 08.00-08.30. The first period indicates the verge of the Wisconsin result (which would be conclusive), where there is actually not much news to share and people probably were just too intense or too tired to share anything. Figure 3., on the next page, visualizes the stream of tweets over the entire election night.

Screen Shot 2017-01-12 at 13.56.52.png

Figure 3. Visualization of the tweets during the Election NIght per 30 minutes interval

Between 07.00-07.30, the most shared URL belonged to a photo of Trump’s office celebrating along with the quote “,” which apparently reflected one of the major fears of Hillary Clinton supporters regarding a Trump presidency (See Appendix C). This tweet came right after AP calling Trump as the winner of Pennsylvania vote, which was the penultimate stop in Trump’s victory before Wisconsin. Until 08.00, people mostly shared live blogs of media outlets, such as BBC, CNN and Washington Post, since the developments accelerated to call the Trump victory conclusively.

Between 08.00-08.30 seems to be a period of transition where Americans go to sleep and international, mainly European, users pick up sharing links (although not confirmed, as there was not enough geolocation data), since URLs from foreign-language media outlets and UK news outlets (BBC has been a globally popular news source throughout the night, but others like Independent were also added to the list at this particular period) get shared. Meanwhile, the main focus shifted to global reactions to Trump’s victory. Figure 4. shows the timeline of URLs by categories.

Figure 4. Timeline of URLs by categories

Another interesting finding was that certain media outlets at opposite ends of polarization, such as Breitbart or Mother Jones did not feature among the most shared URL’s in any period during the night. Mainstream outlets such as CNN or NBC were among the most shared as Figure 5 shows. Figure 5. illustrates the URL’s on a timeline by hosts.

Figure 5. Timeline of urls by hosts

The general research question “How can the U.S. election night 2016 be reconstructed, based on the transition from sharing facts (such as amount of votes) into sharing emotions on Twitter through tweets, words, and links? ” can be answered as following:

Researching the URL’s indicate that, first, period media outlets are shared on twitter, announcing the new president. Second, following tweets contain primal emotions, as a reaction to the result. Tweets contain extreme emotions, positive, negative, and curious about the future. An in-depth study of the negative emotions show the emotional spheres of desperation, angry-disbelieve, disgust, and rebellious-revolting, and fear.

6. Discussion

The US presidential election campaign period, as once a Trump supporter claimed on Full Frontal with Samantha Bee, turned into a reality show. After two years of campaigning rife with polarization and emotions; the Election Night would not be anything different. The surprise win of Donald Trump triggered an outpour of different and striking emotions, a major part of which could be observed on Twitter.

Our triangulated study showed that a great number of people walk into a period of unknowns and concerns, in the United States or elsewhere. Trump’s victory, as expected, seems to have piled on the political polarization that is open to a very emotionalized period of politics. Twitter and other online platforms, in the near future, will be media where this new social fact can be observed. While it is a great research endeavor for social scientists, digital methods will be the key to explore these opportunities. The three kinds of analyses to research emotions and sentiments on Twitter are an attempt to portray the feelings of the people. The frequency, salience, and extremity of negative emotions is supported by the results of the Sentiment Analysis, Timeline Emotions and Discourse Analysis: indicating that the elections trigger people’s first, primal thoughts and feelings through tweets. Metaphorically spoken: it’s like a big earthquake, with aftershocks where people are shocked, trembled, and flabbergasted.

The limitations of this research lie amongst others in the studied tweets. First, The URL-study was based on a select time-period 10PM and 6AM (ET) and focussed on just 8-9 November. A bigger time-period and more selecting more days ⎼⎼ for example 10th November ⎼⎼ as the aftermath would create a different dataset. Second, the dataset selected tweets through TCAT with only the following three keywords: ‘Clinton’, ‘Hillary’ and ‘Trump’, off course there are many other keywords during the US election period, such as ‘Donald’ and ‘US election’. Third, the Discourse Analysis is based on a random sample of mere 2856 tweets. During the entire election period, much more tweets were sent.

The contribution of our research to this ongoing phenomenon is to attempt how these socio-political phenomena can be captured on the fly and analyzed by different methods.

7. Conclusions

7.1 First fifteen minutes

The sentiment analysis and timeline emotion analysis indicate that societal themes such as “racism” and “sexism” evoke extreme emotions on twitter, and create as well negative as positive sentiments. We can conclude from the discourse analysis that negative emotions are supported by a range of primal feelings such as angry-disbelieve, disgust, and rebellious- revolting.

7.2 URL-Study

Thus what we have is the aftermath of Trump’s election is an international event, as much as the US election itself. This is an event that receives global engagement from users. The general feeling seems to be best described as: “Now what will happen?” both among US and international users, but probably with different connotations; one with desperation, other with concern and curiosity. With the point of the conclusive Wisconsin result, as starting point, we observed that the first glimpses of desperation among Hillary Clinton supporters, facing the fact that Trump would become the President-elect. The distinct feeling of desperation may be presumed to be more visible among American users, concerning links, the preoccupied feeling among international users is quite visible as well.

7.3 General conclusion

The general conclusion based on our triangulated research is that the U.S. Elections evoke many primal and extreme emotions and sentiments on Twitter; when examining the negative sentiments in the tweets the future of the US is questioned by the feeling of curiosity, mostly with negative desperate sentiments such as angry-disbelieve, disgust, and rebellious revolting.

8. References

Bekafigo, Marija Anna, and Allan McBride. "Who tweets about politics? Political participation of Twitter users during the 2011 gubernatorial elections." Social Science Computer Review (2013): 0894439313490405.

Bouvier, Gwen, (2015). What is a discourse approach to Twitter, Facebook, YouTube and other social media: connecting with other academic fields? Journal of multicultural discourses (10)2: 149-162

Catenaccio C, Cotter C, De Smedt M, Garzone G, Jacobs G, Macgilchrist F, Lams L, Perrin D, Richardson, JE, Van Hout T, & Van Praet, E. (2011) Towards a linguistics of news production. Journal of Pragmatics 43(7): 1843-1852.

Chiluwa, Innocent. (2015). Radicalist discourse: a study of the stances of Nigeria's Boko Haram and Somalia's Al Shabaab on Twitter. Journal of multicultural discourses (10)2: 214-235

Christensen, Christian. “WAVE-RIDING AND HASHTAG-JUMPING: Twitter, Minority ‘third Parties’ and the 2012 US Elections.” Information, Communication & Society 16, no. 5 (June 2013): 646–66. doi:10.1080/1369118X.2013.783609.

Crooks, Andrew, Arie Croitoru, Anthony Stefanidis, and Jacek Radzikowski. "# Earthquake: Twitter as a distributed sensor system." Transactions in GIS 17, no. 1 (2013): 124-147.

Earle, Paul S., Daniel C. Bowden, and Michelle Guy. "Twitter earthquake detection: earthquake monitoring in a social world." Annals of Geophysics 54, no. 6 (2012).

Gayo-Avello, Daniel. “‘ I Wanted to Predict Elections with Twitter and All I Got Was This Lousy Paper’–A Balanced Survey on Election Prediction Using Twitter Data.” arXiv Preprint arXiv:1204.6441, 2012.

Gee JP (2011) An Introduction to Discourse Analysis: Theory and Method. New York, NY: Routledge.

McCombs, M., & Shaw, D. (1972). The agenda-setting function of massmedia. Public Opinion Quarterly, 36(2), 176–187.

Sakaki, Takeshi, Makoto Okazaki, and Yutaka Matsuo. "Earthquake shakes Twitter users: real-time event detection by social sensors." In Proceedings of the 19th international conference on World wide web, pp. 851-860. ACM, 2010.

Van der Meer T.G.L.A. (2016). Public Frame Building. The Role of Source Usage in Times of Crisis. Communication Research.

Van Dijk, T. A. (1995). Opinions and Ideologies in the Press. In: A. Bell & P. Garrett (Ed). Approaches to media discourse (pp.21-63). Oxford, UK: Blackwell Publishing.

Vergeer, Maurice, and Liesbeth Hermans. "Campaigning on Twitter: Microblogging and online social networking as campaign tools in the 2010 general elections in the Netherlands." Journal of Computer‐Mediated Communication 18, no. 4 (2013): 399-419.

Appendix A Word cloud

Screen Shot 2017-01-10 at 12.39.30.png

Appendix B Additional graph of the Discourse Analysis


-- NataliaSanchez - 20 Feb 2017
Topic revision: r1 - 20 Feb 2017, NataliaSanchez
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback