We are big data analyst for a communication agency who wants to analyze Donald Trump’s communication on Twitter.
The text file trump_tweets.txt contain an history of all trump tweets from 2009 to November 19th, 2020.
Using spark we will explore and analyze this data and get vizualization.
Summary
First let's import pyspark and create a SparkContext
import pyspark
sc = pyspark.SparkContext(appName="Trump-Tweets")
We divide our study in two parts analysis based on the text of the tweets then on the date
First we read the trump_tweets.txt file into the RDD rdd_trump_tweets
rdd_trump_tweets = sc.textFile('trump_tweets.txt')
rdd_trump_tweets.take(20)
['Republicans and Democrats have both created our economic problems.;02/08/2011 18:07:48', 'I was thrilled to be back in the Great city of Charlotte, North Carolina with thousands of hardworking American Patriots who love our Country, cherish our values, respect our laws, and always put AMERICA FIRST! Thank you for a wonderful evening!! #KAG2020 https://t.co/dNJZfRsl9y;03/03/2020 01:34:50', 'RT @CBS_Herridge: READ: Letter to surveillance court obtained by CBS News questions where there will be further disciplinary action and cho…;17/01/2020 03:22:47', 'The Unsolicited Mail In Ballot Scam is a major threat to our Democracy, & the Democrats know it. Almost all recent elections using this system, even though much smaller & with far fewer Ballots to count, have ended up being a disaster. Large numbers of missing Ballots & Fraud!;12/09/2020 20:10:58', "RT @MZHemingway: Very friendly telling of events here about Comey's apparent leaking to compliant media. If you read those articles and tho…;17/01/2020 13:13:59", 'RT @WhiteHouse: President @realDonaldTrump announced historic steps to protect the Constitutional right to pray in public schools! https://…;17/01/2020 00:11:56', '“I’m running as a proud Democrat, for the Senate”. Sleepy Joe Biden today. It’s only going to get worse. It is not sustainable for our County - China will own us!!!!;12/10/2020 22:22:39', 'Getting a little exercise this morning! https://t.co/fyAAcbhbgk;01/02/2020 16:14:02', 'https://t.co/4qwCKQOiOw;23/10/2020 04:52:14', 'https://t.co/VlEu8yyovv;23/10/2020 04:46:53', 'https://t.co/z5CRqHO8vg;23/10/2020 04:46:49', 'https://t.co/TQCQiDrVOB;23/10/2020 04:46:45', 'RT @EricTrump: https://t.co/NcrNdOSfIV;23/10/2020 04:30:19', '"RT @GOPChairwoman: The economic boom continues!', '', '273K jobs added in February ', '', '7M+ jobs added since @realDonaldTrump was elected', '', 'Unemploym…";08/03/2020 04:01:46']
We split the lines to isolate the text from the date
rdd_trump_text = rdd_trump_tweets.map(lambda tweet: tweet.split(';')[0])
rdd_trump_text.take(20)
['Republicans and Democrats have both created our economic problems.', 'I was thrilled to be back in the Great city of Charlotte, North Carolina with thousands of hardworking American Patriots who love our Country, cherish our values, respect our laws, and always put AMERICA FIRST! Thank you for a wonderful evening!! #KAG2020 https://t.co/dNJZfRsl9y', 'RT @CBS_Herridge: READ: Letter to surveillance court obtained by CBS News questions where there will be further disciplinary action and cho…', 'The Unsolicited Mail In Ballot Scam is a major threat to our Democracy, & the Democrats know it. Almost all recent elections using this system, even though much smaller & with far fewer Ballots to count, have ended up being a disaster. Large numbers of missing Ballots & Fraud!', "RT @MZHemingway: Very friendly telling of events here about Comey's apparent leaking to compliant media. If you read those articles and tho…", 'RT @WhiteHouse: President @realDonaldTrump announced historic steps to protect the Constitutional right to pray in public schools! https://…', '“I’m running as a proud Democrat, for the Senate”. Sleepy Joe Biden today. It’s only going to get worse. It is not sustainable for our County - China will own us!!!!', 'Getting a little exercise this morning! https://t.co/fyAAcbhbgk', 'https://t.co/4qwCKQOiOw', 'https://t.co/VlEu8yyovv', 'https://t.co/z5CRqHO8vg', 'https://t.co/TQCQiDrVOB', 'RT @EricTrump: https://t.co/NcrNdOSfIV', '"RT @GOPChairwoman: The economic boom continues!', '', '273K jobs added in February ', '', '7M+ jobs added since @realDonaldTrump was elected', '', 'Unemploym…"']
We split the text into words to run our sentiment analysis
rdd_trump_text_words = rdd_trump_text.flatMap(lambda line:line.split())\
.flatMap(lambda line:line.split('.')).flatMap(lambda line:line.split(','))\
.flatMap(lambda line:line.split('"')).flatMap(lambda line:line.split('”'))\
.flatMap(lambda line:line.split('?')).flatMap(lambda line:line.split('!'))\
.flatMap(lambda line:line.split(':'))\
.filter(lambda word : word != "").map(lambda word: word.lower())
rdd_trump_text_words.take(20)
['republicans', 'and', 'democrats', 'have', 'both', 'created', 'our', 'economic', 'problems', 'i', 'was', 'thrilled', 'to', 'be', 'back', 'in', 'the', 'great', 'city', 'of']
Here we create our own list of stop words: some words with no sense doesn't have to be part of the sentiment analysis
list_our_stop_words = ['http','https','//t','rt','@realdonaldtrump','&','u','s']
rdd_our_stop_words = sc.parallelize(list_our_stop_words)
rdd_our_stop_words.take(5)
['http', 'https', '//t', 'rt', '@realdonaldtrump']
rdd_stop_words = sc.textFile('stop-words.txt')
rdd_stop_words.take(10)
['a', 'about', 'above', 'across', 'after', 'afterwards', 'again', 'against', 'all', 'almost']
rdd_stop_words_kv = rdd_stop_words.map(lambda word: (word,1))
rdd_stop_words_kv.take(10)
[('a', 1),
('about', 1),
('above', 1),
('across', 1),
('after', 1),
('afterwards', 1),
('again', 1),
('against', 1),
('all', 1),
('almost', 1)]
rdd_negative_words = sc.textFile('negative-words.txt')
rdd_negative_words.take(10)
['2-faced', '2-faces', 'abnormal', 'abolish', 'abominable', 'abominably', 'abominate', 'abomination', 'abort', 'aborted']
rdd_negative_words_kv = rdd_negative_words.map(lambda word: (word,1))
rdd_negative_words_kv.take(10)
[('2-faced', 1),
('2-faces', 1),
('abnormal', 1),
('abolish', 1),
('abominable', 1),
('abominably', 1),
('abominate', 1),
('abomination', 1),
('abort', 1),
('aborted', 1)]
rdd_positive_words = sc.textFile('positive-words.txt')
rdd_positive_words.take(20)
['a+', 'abound', 'abounds', 'abundance', 'abundant', 'accessable', 'accessible', 'acclaim', 'acclaimed', 'acclamation', 'accolade', 'accolades', 'accommodative', 'accomodative', 'accomplish', 'accomplished', 'accomplishment', 'accomplishments', 'accurate', 'accurately']
The word "trump" is included in the positive words and, as it's the name of the president, it must be removed from the list so not to distort our study because it's one of the most used words in his tweets and retweets
list_positive_words_to_remove = ['trump']
rdd_positive_words_to_remove = sc.parallelize(list_positive_words_to_remove)
rdd_positive_words = rdd_positive_words.subtract(rdd_positive_words_to_remove)
rdd_positive_words_kv = rdd_positive_words.map(lambda word: (word,1))
rdd_positive_words_kv.take(20)
[('abound', 1),
('accessable', 1),
('accessible', 1),
('accomplishment', 1),
('achievements', 1),
('acumen', 1),
('adroit', 1),
('adroitly', 1),
('adulatory', 1),
('advantageously', 1),
('affirmation', 1),
('afordable', 1),
('agilely', 1),
('all-around', 1),
('ambitiously', 1),
('amply', 1),
('appealing', 1),
('appreciates', 1),
('astoundingly', 1),
('attune', 1)]
rdd_trump_text_words_kv = rdd_trump_text_words.map(lambda word: (word,1))
rdd_trump_text_words_kv.take(20)
[('republicans', 1),
('and', 1),
('democrats', 1),
('have', 1),
('both', 1),
('created', 1),
('our', 1),
('economic', 1),
('problems', 1),
('i', 1),
('was', 1),
('thrilled', 1),
('to', 1),
('be', 1),
('back', 1),
('in', 1),
('the', 1),
('great', 1),
('city', 1),
('of', 1)]
We join the positive words and words from the tweets
rdd_trump_text_words_positive = rdd_positive_words_kv.join(rdd_trump_text_words_kv).mapValues(lambda tuple: 1)
rdd_trump_text_words_positive.take(10)
[('authentic', 1),
('booming', 1),
('booming', 1),
('booming', 1),
('booming', 1),
('booming', 1),
('booming', 1),
('booming', 1),
('booming', 1),
('booming', 1)]
We reduce by key and sort by descending order based on the count to get the occurrences of each positive word in the tweets
rdd_trump_text_words_positive_count_ordered = rdd_trump_text_words_positive.reduceByKey(lambda x,y : x+y)\
.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_positives_opinion = rdd_trump_text_words_positive_count_ordered.take(20)
list_top_positives_opinion
[('great', 7464),
('thank', 3525),
('like', 2005),
('good', 1816),
('best', 1234),
('love', 1160),
('win', 1118),
('work', 1014),
('right', 934),
('better', 851),
('well', 757),
('strong', 734),
('support', 673),
('amazing', 647),
('congratulations', 598),
('nice', 580),
('happy', 577),
('honor', 500),
('wow', 495),
('wonderful', 482)]
We follow the same procedure for negative words too
rdd_trump_text_words_negative = rdd_negative_words_kv.join(rdd_trump_text_words_kv).mapValues(lambda tuple: 1)
rdd_trump_text_words_negative.take(10)
[('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1)]
rdd_trump_text_words_negative_count_ordered = rdd_trump_text_words_negative.reduceByKey(lambda x,y : x+y)\
.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_negatives_opinion = rdd_trump_text_words_negative_count_ordered.take(20)
list_top_negatives_opinion
[('fake', 1214),
('bad', 999),
('hard', 751),
('crime', 554),
('wrong', 449),
('crooked', 447),
('illegal', 420),
('radical', 385),
('terrible', 384),
('lost', 369),
('corrupt', 354),
('collusion', 349),
('sad', 330),
('disaster', 310),
('problem', 302),
('hoax', 297),
('crazy', 290),
('attack', 283),
('miss', 277),
('failed', 267)]
We create a function to plot the positive and negative words
from matplotlib import pyplot as plt
import seaborn as sb
import pandas as pd
def plot_opinion(list_positives, list_negatives):
list_1 = [(t[0], t[1], 'positive') for t in list_positives]
list_2 = [(t[0], t[1], 'negative') for t in list_negatives]
list_1.extend(list_2)
pd_df = pd.DataFrame({
'Opinion': [ t[0] for t in list_1 ],
'Count': [ t[1] for t in list_1 ],
'polarity': [ t[2] for t in list_1 ]
}).sort_values('Count', ascending=False)
pd_df['color'] = pd_df.polarity.apply(lambda polarity: 'green' if polarity=='positive' else 'red')
plt.figure( figsize = (15, 8) )
chart=sb.barplot(data=pd_df, x='Opinion', y='Count', hue='polarity', palette=["green", "red"], order=pd_df['Opinion'])
chart.set_xticklabels(
chart.get_xticklabels(),
rotation=80,
horizontalalignment='right',
fontweight='light',
fontsize='x-large',
)
for tick, color in zip(chart.get_xticklabels(), pd_df['color']):
tick.set_color(color)
plt.show()
plot_opinion(list_top_positives_opinion, list_top_negatives_opinion)
We can already see from this graph that Trump uses far more positive words in his tweets than negative ones, so we can expect a positive sentiment score. This might be surprising as Trump is seen as a very nervous president and we might expect a negative sentiment score: however, he probably uses a different tone on Twitter than we see him on TV, so we will probably have a positive score.
We compute the global sentiment score: the difference between the number of positive words and the number of negative words
sentiment_score = rdd_trump_text_words_positive.map(lambda tuple: tuple[1]).sum()\
- rdd_trump_text_words_negative.map(lambda tuple: tuple[1]).sum()
print('The sentiment score is : ',sentiment_score)
The sentiment score is : 19355
Our previous assumptions were confirmed with a rather high sentiment score
We study and count the occurrence of words that are neither negative nor positive and are called contextual words
We remove all the positive, negative and stop words from the tweets words to keep only the contextual words
rdd_trump_text_contextual_words = rdd_trump_text_words.subtract(rdd_positive_words).subtract(rdd_negative_words).subtract(rdd_stop_words).subtract(rdd_our_stop_words).subtract(rdd_positive_words_to_remove)
rdd_trump_text_contextual_words.take(20)
['co/ai2vtb1v3g', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends']
We assign on each word a value of 1 and reduce by key to obtain the occurences of contextual words, sorted by descending order
rdd_trump_text_contextual_words_ordered = rdd_trump_text_contextual_words.map(lambda word: (word, 1))\
.reduceByKey(lambda x,y : x+y)\
.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_contextual_words = rdd_trump_text_contextual_words_ordered.take(50)
list_top_contextual_words
[('president', 4501),
('people', 3365),
('just', 3280),
('new', 2477),
('country', 2255),
('thanks', 2183),
('america', 2170),
('big', 2054),
('time', 1887),
('make', 1835),
('news', 1763),
('donald', 1713),
('democrats', 1694),
('vote', 1658),
('obama', 1651),
('today', 1628),
('going', 1521),
('want', 1409),
('american', 1400),
('years', 1266),
('job', 1255),
('run', 1254),
('need', 1204),
('doing', 1175),
('media', 1128),
('think', 1123),
('china', 1122),
('way', 1098),
('really', 1095),
('did', 1079),
('know', 1075),
('day', 1071),
('house', 1049),
('jobs', 1027),
('said', 1005),
('states', 985),
('true', 983),
('election', 973),
('border', 968),
('deal', 951),
('state', 946),
('world', 939),
('@foxnews', 927),
('biden', 910),
('hillary', 903),
("don't", 902),
('tonight', 900),
('watch', 898),
('united', 868),
('@whitehouse', 836)]
We create a function to plot the contextual words
def plot_contextual_words(list_contextual_words):
pd_df = pd.DataFrame({
'Word': [ t[0] for t in list_contextual_words ],
'Count': [ t[1] for t in list_contextual_words ]
}).sort_values('Count', ascending=False)
plt.figure( figsize = (15, 8) )
chart=sb.barplot(data=pd_df, x='Word', y='Count', order=pd_df['Word'], color='blue')
chart.set_xticklabels(
chart.get_xticklabels(),
rotation=80,
horizontalalignment='right',
fontweight='light',
fontsize='x-large',
)
plt.show()
plot_contextual_words(list_top_contextual_words)
From the words most used by Trump, we can see that his speeches are mainly focused on politics (as we expected): frequently used words are "president", "people", "America", "democrats", "votes", "Obama", "China", "border", "Biden"....
We use a wordcloud to get another visualization
#pip install wordcloud
from wordcloud import WordCloud
import matplotlib.pyplot as plt
import seaborn as sb
import pandas as pd
def plot_cloud(list_):
wordcloud = WordCloud(width = 1500, height = 500, random_state=1, background_color='white', colormap='Set2', collocations=False).generate_from_frequencies(dict(list_))
# Set figure size
plt.figure(figsize=(40, 30))
# Display image
plt.imshow(wordcloud)
# No axis details
plt.axis("off")
plt.show();
plot_cloud(list_top_contextual_words)
We wanted to focus our study on what seemed to us to be hot topics, namely covid-19 then the presidential elections.
We filter and keep only the tweets containing the word covid
rdd_trump_text_covid = rdd_trump_text.map(lambda word: word.lower()).filter(lambda line:'covid' in line)
rdd_trump_text_covid.take(20)
['joe biden’s response to the h1n1 swine flu, far less lethal than covid 19, was one of the weakest and worst in the history of fighting epidemics and pandemics. it was pathetic, those involved have said. joe didn’t have a clue!', 'rt @drmarcsiegel: my new book separates fact from fiction in the covid-19 epidemic. covid: the politics of fear and the power of science is…', 'rt @therightmelissa: @realdonaldtrump the nerve of gov. cuomo trying to capitalize on covid after he killed 1000’s in the nursing homes?? n…', 'governor cuomo has shown tremendously poor leadership skills in running n.y. bad time for him to be writing and promoting a book, especially since he has done such a poor job with covid and crime. so many unnecessary deaths. the city & state have sunk to historic lows...', 'the great bobby bowden, one of the best coaches ever in college football, on his full recovery from covid-19 “....america is the greatest country this side of heaven. i’ve had a chance to get a lot of wins in my life, but i really wanted to win this one because i wanted....', 'we have made tremendous progress with the china virus, but the fake news refuses to talk about it this close to the election. covid, covid, covid is being used by them, in total coordination, in order to change our great early election numbers.should be an election law violation!', 'the fake news media is riding covid, covid, covid, all the way to the election. losers!', 'rt @realdonaldtrump: the fake news media is riding covid, covid, covid, all the way to the election. losers!', 'all the fake news media wants to talk about is covid, covid, covid. on november 4th, you won’t be hearing so much about it anymore. we are rounding the turn!!!', 'until november 4th., fake news media is going full on covid, covid, covid. we are rounding the turn. 99.9%.', 'covid, covid, covid is the unified chant of the fake news lamestream media. they will talk about nothing else until november 4th., when the election will be (hopefully!) over. then the talk will be how low the death rate is, plenty of hospital rooms, & many tests of young people.', 'topic: #covid19nc', 'hope hicks, who has been working so hard without even taking a small break, has just tested positive for covid 19. terrible! the first lady and i are waiting for our test results. in the meantime, we will begin our quarantine process!', 'rt @whitehouse: president @realdonaldtrump is working to ensure that america produces the critical goods necessary to combat covid-19 here…', 'in an illegal late night coup, nevada’s clubhouse governor made it impossible for republicans to win the state. post office could never handle the traffic of mail-in votes without preparation. using covid to steal the state. see you in court! https://t.co/cnspingcy7', 'the @nytimes now admits that patients ill with covid are getting better faster and at higher (much) percentages.', 'biden did a very poor job of handling h1n1 swine flu, yet all he talks about is covid, covid, covid, like he would actually know what he is doing. he doesn’t. plus, he would be terrible with the economy!', 'rt @mitchellvii: the economy is going to do so well this summer and make trump look so good that democrats will accuse him of hyping covid-…', 'rt @hhsgov: watch this video to learn how you can help prevent the spread of respiratory diseases like the novel #coronavirus (#covid19). t…', 'rt @cdcgov: everyone can help prevent the spread of #covid19. call your doctor if you develop symptoms, have been in close contact with a p…']
rdd_trump_text_covid_words = rdd_trump_text_covid.flatMap(lambda line:line.split())\
.flatMap(lambda line:line.split('.')).flatMap(lambda line:line.split(','))\
.flatMap(lambda line:line.split('"')).flatMap(lambda line:line.split('”'))\
.flatMap(lambda line:line.split('?')).flatMap(lambda line:line.split('!'))\
.flatMap(lambda line:line.split(':'))\
.filter(lambda word : word != "")
rdd_trump_text_covid_words.take(20)
['joe', 'biden’s', 'response', 'to', 'the', 'h1n1', 'swine', 'flu', 'far', 'less', 'lethal', 'than', 'covid', '19', 'was', 'one', 'of', 'the', 'weakest', 'and']
rdd_trump_text_covid_words_kv = rdd_trump_text_covid_words.map(lambda word: (word,1))
rdd_trump_text_covid_words_kv.take(20)
[('joe', 1),
('biden’s', 1),
('response', 1),
('to', 1),
('the', 1),
('h1n1', 1),
('swine', 1),
('flu', 1),
('far', 1),
('less', 1),
('lethal', 1),
('than', 1),
('covid', 1),
('19', 1),
('was', 1),
('one', 1),
('of', 1),
('the', 1),
('weakest', 1),
('and', 1)]
Then we join the tweets words with positive and negative words
rdd_trump_text_covid_words_positive = rdd_positive_words_kv.join(rdd_trump_text_covid_words_kv).mapValues(lambda tuple: 1)
rdd_trump_text_covid_words_positive.take(10)
[('free', 1),
('right', 1),
('right', 1),
('advantages', 1),
('proactive', 1),
('grateful', 1),
('grateful', 1),
('like', 1),
('like', 1),
('like', 1)]
rdd_trump_text_covid_words_positive_count = rdd_trump_text_covid_words_positive.reduceByKey(lambda x,y : x+y)
rdd_trump_text_covid_words_positive_count.take(20)
[('free', 1),
('right', 2),
('advantages', 1),
('proactive', 1),
('grateful', 2),
('like', 5),
('pleased', 1),
('wonderful', 1),
('faster', 1),
('confidence', 1),
('encouraging', 3),
('works', 1),
('stable', 1),
('honor', 1),
('significant', 2),
('tremendously', 1),
('helping', 1),
('positive', 5),
('available', 2),
('enough', 1)]
rdd_trump_text_covid_words_positive_count_ordered = rdd_trump_text_covid_words_positive_count.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_covid_positive_opinion = rdd_trump_text_covid_words_positive_count_ordered.take(20)
list_top_covid_positive_opinion
[('great', 13),
('support', 8),
('good', 7),
('thank', 6),
('like', 5),
('positive', 5),
('work', 5),
('strong', 4),
('relief', 4),
('encouraging', 3),
('important', 3),
('effective', 3),
('win', 3),
('protect', 3),
('better', 3),
('right', 2),
('grateful', 2),
('significant', 2),
('available', 2),
('well', 2)]
rdd_trump_text_covid_words_negative = rdd_negative_words_kv.join(rdd_trump_text_covid_words_kv).mapValues(lambda tuple: 1)
rdd_trump_text_covid_words_negative.take(10)
[('absentee', 1),
('confusion', 1),
('disaster', 1),
('drastically', 1),
('emergency', 1),
('emergency', 1),
('emergency', 1),
('epidemic', 1),
('fraud', 1),
('gross', 1)]
rdd_trump_text_covid_words_negative_count = rdd_trump_text_covid_words_negative.reduceByKey(lambda x,y : x+y)
rdd_trump_text_covid_words_negative_count.take(20)
[('absentee', 1),
('confusion', 1),
('disaster', 1),
('drastically', 1),
('emergency', 3),
('epidemic', 1),
('fraud', 1),
('gross', 1),
('horrible', 1),
('illegal', 1),
('impossible', 1),
('infected', 1),
('killed', 1),
('loss', 1),
('lost', 2),
('mistake', 1),
('pathetic', 1),
('poorly', 1),
('rejecting', 1),
('risk', 4)]
rdd_trump_text_covid_words_negative_count_ordered = rdd_trump_text_covid_words_negative_count.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_covid_negative_opinion = rdd_trump_text_covid_words_negative_count_ordered.take(20)
list_top_covid_negative_opinion
[('fake', 12),
('death', 6),
('outbreak', 6),
('slow', 5),
('risk', 4),
('hard', 4),
('symptoms', 4),
('virus', 4),
('emergency', 3),
('breaking', 3),
('crisis', 3),
('poor', 3),
('problem', 3),
('lost', 2),
('bad', 2),
('break', 2),
('illness', 2),
('lies', 2),
('losers', 2),
('outrageous', 2)]
We create a function to plot the positive and negative words
from matplotlib import pyplot as plt
import seaborn as sb
import pandas as pd
def plot_opinion_covid(list_positives, list_negatives):
list_1 = [(t[0], t[1], 'positive') for t in list_positives]
list_2 = [(t[0], t[1], 'negative') for t in list_negatives]
list_1.extend(list_2)
pd_df = pd.DataFrame({
'Opinion': [ t[0] for t in list_1 ],
'Count': [ t[1] for t in list_1 ],
'polarity': [ t[2] for t in list_1 ]
}).sort_values('Count', ascending=False)
pd_df['color'] = pd_df.polarity.apply(lambda polarity: 'green' if polarity=='positive' else 'red')
plt.figure( figsize = (15, 8) )
chart=sb.barplot(data=pd_df, x='Opinion', y='Count', hue='polarity', palette=["green", "red"], order=pd_df['Opinion'])
chart.set_xticklabels(
chart.get_xticklabels(),
rotation=80,
horizontalalignment='right',
fontweight='light',
fontsize='x-large',
)
for tick, color in zip(chart.get_xticklabels(), pd_df['color']):
tick.set_color(color)
plt.show()
plot_opinion_covid(list_top_covid_positive_opinion, list_top_covid_negative_opinion)
In this graph we can see that there is a higher number of negative words, which is understandable in view of the subject. It would seem that President Trump is less positive and he is worried in relation to the topic of Covid-19: very negative words are used in his tweets like "death", "virus", "hard", "risk", "emergency", "crisis".
sentiment_score_covid = rdd_trump_text_covid_words_positive_count.map(lambda tuple: tuple[1]).sum()\
- rdd_trump_text_covid_words_negative_count.map(lambda tuple: tuple[1]).sum()
print('The sentiment score is : ',sentiment_score_covid)
The sentiment score is : -13
Our previous assumptions were confirmed with a more neutral sentiment score.
We create a list of words that need to be removed from our studies. These words are linked to the word covid and are of no interest.
list_our_stop_words_covid = ['#covid19','covid','covid-19','19']
rdd_our_stop_words_covid = sc.parallelize(list_our_stop_words_covid)
rdd_our_stop_words_covid.take(5)
['#covid19', 'covid', 'covid-19', '19']
rdd_trump_text_covid_contextual_words = rdd_trump_text_covid_words.subtract(rdd_positive_words).subtract(rdd_negative_words).subtract(rdd_stop_words).subtract(rdd_our_stop_words).subtract(rdd_our_stop_words_covid)
rdd_trump_text_covid_contextual_words.take(20)
['bobby', 'narrative', 'narrative', '@foxandfriends', 'tre…', 'guest', 'start', 'dod’s', 'family', '(and', '“to', 'texas', 'texas', 'utilize', 'maintain…', 'guard', 'los', 'com…', '#clinicaltrial', '#clinicaltrial']
rdd_trump_text_covid_contextual_words_ordered = rdd_trump_text_covid_contextual_words.map(lambda word: (word, 1))\
.reduceByKey(lambda x,y : x+y)\
.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_contextual_covid_words = rdd_trump_text_covid_contextual_words_ordered.take(20)
list_top_contextual_covid_words
[('new', 18),
('@cdcgov', 18),
('news', 16),
('people', 15),
('spread', 14),
('help', 14),
('media', 13),
('response', 10),
('just', 10),
('today', 9),
('vaccine', 9),
('election', 8),
('testing', 8),
('trump', 8),
('000', 8),
('states', 8),
('cases', 7),
('covid19', 7),
('working', 7),
('@whitehouse', 7)]
We create a function to plot the most common contextual words
def plot_contextual_covid_words(list_contextual_words):
pd_df = pd.DataFrame({
'Word': [ t[0] for t in list_contextual_words ],
'Count': [ t[1] for t in list_contextual_words ]
}).sort_values('Count', ascending=False)
plt.figure( figsize = (15, 8) )
chart=sb.barplot(data=pd_df, x='Word', y='Count', order=pd_df['Word'], color='blue')
chart.set_xticklabels(
chart.get_xticklabels(),
rotation=80,
horizontalalignment='right',
fontweight='light',
fontsize='x-large',
)
plt.show()
plot_contextual_covid_words(list_top_contextual_covid_words)
All the contextual words are related to Covid-19, the health emergency, the possible vaccine and also the influence the pandemic may have on the political world (elections).
We filter and keep only the tweets containing the word election
rdd_trump_text_election = rdd_trump_text.map(lambda word: word.lower()).filter(lambda line:'election' in line)
rdd_trump_text_election.take(10)
['the unsolicited mail in ballot scam is a major threat to our democracy, & the democrats know it. almost all recent elections using this system, even though much smaller & with far fewer ballots to count, have ended up being a disaster. large numbers of missing ballots & fraud!', 'rt @chefjclark: people ask why the president fights so hard in court for ballot integrity? why we have such an extensive election day oper…', 'what a terrible thing for biden to say! rigged election? https://t.co/jqxxeqczu9', 'this is the most important election in the history of our country. with your vote, we will continue to bring back your jobs, lower drug prices, support our police, protect the #2a, defend our borders, and ensure more products are proudly stamped with the phrase made in the usa! https://t.co/m0ba8qr7nl', 'rt @texasgop: report any irregularities to the republican party of texas voter helpline: 855-433-1663 #earlyvoting #vote #election2020 http…', '“a pennsylvania appeals court turns down the trump administrations request for poll watchers to monitor satellite election offices.” @oann how terrible is this? we are just seeking a fair vote count. this can only lead to very bad things. bad intentions much??? disgraceful!!!', 'happening all over usa. no wonder california is never in play for republicans. rigged election! https://t.co/abfieyy0kn', 'we have made tremendous progress with the china virus, but the fake news refuses to talk about it this close to the election. covid, covid, covid is being used by them, in total coordination, in order to change our great early election numbers.should be an election law violation!', 'the fake news media is riding covid, covid, covid, all the way to the election. losers!', 'rt @realdonaldtrump: the fake news media is riding covid, covid, covid, all the way to the election. losers!']
rdd_trump_text_election_words = rdd_trump_text_election.flatMap(lambda line:line.split())\
.flatMap(lambda line:line.split('.')).flatMap(lambda line:line.split(','))\
.flatMap(lambda line:line.split('"')).flatMap(lambda line:line.split('”'))\
.flatMap(lambda line:line.split('?')).flatMap(lambda line:line.split('!'))\
.flatMap(lambda line:line.split(':'))\
.filter(lambda word : word != "")
rdd_trump_text_election_words.take(20)
['the', 'unsolicited', 'mail', 'in', 'ballot', 'scam', 'is', 'a', 'major', 'threat', 'to', 'our', 'democracy', '&', 'the', 'democrats', 'know', 'it', 'almost', 'all']
rdd_trump_text_election_words_kv = rdd_trump_text_election_words.map(lambda word: (word,1))
rdd_trump_text_election_words_kv.take(20)
[('the', 1),
('unsolicited', 1),
('mail', 1),
('in', 1),
('ballot', 1),
('scam', 1),
('is', 1),
('a', 1),
('major', 1),
('threat', 1),
('to', 1),
('our', 1),
('democracy', 1),
('&', 1),
('the', 1),
('democrats', 1),
('know', 1),
('it', 1),
('almost', 1),
('all', 1)]
rdd_trump_text_election_words_positive = rdd_positive_words_kv.join(rdd_trump_text_election_words_kv).mapValues(lambda tuple: 1)
rdd_trump_text_election_words_positive.take(10)
[('booming', 1),
('booming', 1),
('free', 1),
('free', 1),
('free', 1),
('free', 1),
('led', 1),
('right', 1),
('right', 1),
('right', 1)]
rdd_trump_text_election_words_positive_count = rdd_trump_text_election_words_positive.reduceByKey(lambda x,y : x+y)
rdd_trump_text_election_words_positive_count.take(20)
[('booming', 2),
('free', 4),
('led', 1),
('right', 21),
('willing', 3),
('leading', 5),
('smart', 7),
('admirable', 1),
('conveniently', 2),
('excellent', 2),
('capable', 1),
('grateful', 2),
('like', 40),
('liked', 3),
('pleased', 1),
('safe', 6),
('stunning', 1),
('competitive', 2),
('fearless', 2),
('genius', 1)]
rdd_trump_text_election_words_positive_count_ordered = rdd_trump_text_election_words_positive_count.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_election_positive_opinion = rdd_trump_text_election_words_positive_count_ordered.take(20)
list_top_election_positive_opinion
[('great', 122),
('win', 113),
('won', 61),
('like', 40),
('important', 32),
('wins', 32),
('strong', 29),
('thank', 29),
('good', 27),
('best', 24),
('support', 22),
('right', 21),
('well', 21),
('better', 19),
('victory', 17),
('work', 17),
('tough', 16),
('winning', 16),
('greatest', 16),
('lead', 15)]
rdd_trump_text_election_words_negative = rdd_negative_words_kv.join(rdd_trump_text_election_words_kv).mapValues(lambda tuple: 1)
rdd_trump_text_election_words_negative.take(10)
[('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1),
('absentee', 1)]
rdd_trump_text_election_words_negative_count = rdd_trump_text_election_words_negative.reduceByKey(lambda x,y : x+y)
rdd_trump_text_election_words_negative_count.take(20)
[('absentee', 11),
('anti-us', 1),
('bizarre', 1),
('blatant', 1),
('chaos', 2),
('concede', 1),
('conceded', 3),
('concerns', 2),
('confusion', 1),
('conservative', 1),
('costly', 1),
('crazy', 10),
('crushing', 1),
('denying', 1),
('detrimental', 1),
('disaster', 14),
('discredit', 1),
('disgrace', 1),
('dishonest', 4),
('dismal', 1)]
rdd_trump_text_election_words_negative_count_ordered = rdd_trump_text_election_words_negative_count.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_election_negative_opinion = rdd_trump_text_election_words_negative_count_ordered.take(20)
list_top_election_negative_opinion
[('fake', 65),
('fraud', 33),
('bad', 32),
('crooked', 31),
('lost', 29),
('crime', 28),
('hard', 24),
('wrong', 23),
('collusion', 20),
('corrupt', 18),
('excuse', 18),
('steal', 18),
('illegal', 17),
('breaking', 15),
('phony', 15),
('disaster', 14),
('lose', 14),
('hoax', 14),
('radical', 14),
('illegally', 13)]
We create a function to plot the positive and negative words
from matplotlib import pyplot as plt
import seaborn as sb
import pandas as pd
def plot_opinion_election(list_positives, list_negatives):
list_1 = [(t[0], t[1], 'positive') for t in list_positives]
list_2 = [(t[0], t[1], 'negative') for t in list_negatives]
list_1.extend(list_2)
pd_df = pd.DataFrame({
'Opinion': [ t[0] for t in list_1 ],
'Count': [ t[1] for t in list_1 ],
'polarity': [ t[2] for t in list_1 ]
}).sort_values('Count', ascending=False)
pd_df['color'] = pd_df.polarity.apply(lambda polarity: 'green' if polarity=='positive' else 'red')
plt.figure( figsize = (15, 8) )
chart=sb.barplot(data=pd_df, x='Opinion', y='Count', hue='polarity', palette=["green", "red"], order=pd_df['Opinion'])
chart.set_xticklabels(
chart.get_xticklabels(),
rotation=80,
horizontalalignment='right',
fontweight='light',
fontsize='x-large',
)
for tick, color in zip(chart.get_xticklabels(), pd_df['color']):
tick.set_color(color)
plt.show()
plot_opinion_election(list_top_election_positive_opinion, list_top_election_negative_opinion)
There is a mix of positive and negative words related to the election, Trump does not have a firm opinion about it and most importantly a more thorough study should be done to distinguish the 2016 and 2020 elections.
sentiment_score_election = rdd_trump_text_election_words_positive_count.map(lambda tuple: tuple[1]).sum()\
- rdd_trump_text_election_words_negative_count.map(lambda tuple: tuple[1]).sum()
print('The sentiment score is : ',sentiment_score_election)
The sentiment score is : -62
list_our_stop_words_election = ['election']
rdd_our_stop_words_election = sc.parallelize(list_our_stop_words_election)
rdd_our_stop_words_election.take(5)
['election']
rdd_trump_text_election_contextual_words = rdd_trump_text_election_words.subtract(rdd_positive_words).subtract(rdd_negative_words).subtract(rdd_stop_words).subtract(rdd_our_stop_words).subtract(rdd_our_stop_words_election)
rdd_trump_text_election_contextual_words.take(20)
['24', '24', '24', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', '@foxandfriends', 'caight', '(we', '@zedd_tahlz', '@the_lady_colors', 'narrative', 'narrative', 'narrative']
rdd_trump_text_election_contextual_words_ordered = rdd_trump_text_election_contextual_words.map(lambda word: (word, 1))\
.reduceByKey(lambda x,y : x+y)\
.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_contextual_election_words = rdd_trump_text_election_contextual_words_ordered.take(20)
list_top_contextual_election_words
[('trump', 147),
('2016', 117),
('democrats', 112),
('president', 112),
('vote', 102),
('just', 101),
('people', 94),
('obama', 82),
('big', 77),
('2020', 76),
('news', 74),
('day', 73),
('elections', 72),
('rigged', 64),
('did', 59),
('ballots', 58),
('going', 57),
('country', 56),
('new', 53),
('want', 53)]
We create a function to plot the most common contextual words related to the topic election.
def plot_contextual_election_words(list_contextual_words):
pd_df = pd.DataFrame({
'Word': [ t[0] for t in list_contextual_words ],
'Count': [ t[1] for t in list_contextual_words ]
}).sort_values('Count', ascending=False)
plt.figure( figsize = (15, 8) )
chart=sb.barplot(data=pd_df, x='Word', y='Count', order=pd_df['Word'], color='blue')
chart.set_xticklabels(
chart.get_xticklabels(),
rotation=80,
horizontalalignment='right',
fontweight='light',
fontsize='x-large',
)
plt.show()
plot_contextual_election_words(list_top_contextual_election_words)
This graph shows that Trump mentioned the 2016 election more than the 2020 election, probably his campaigning was different in the two elections and influenced the final result. He also had to make himself known more in 2016 than 2020.
We make an analysis of the most used hashtags in Trump's tweets
rdd_hashtags = rdd_trump_text_words.filter(lambda line : line.startswith('#'))\
.map(lambda word: (word,1))\
.reduceByKey(lambda x,y : x+y)\
.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_hashtags = rdd_hashtags.take(30)
list_top_hashtags
[('#trump2016', 799),
('#makeamericagreatagain', 531),
('#maga', 475),
('#celebapprentice', 297),
('#celebrityapprentice', 150),
('#1', 141),
('#trump', 121),
('#timetogettough', 100),
('#americafirst', 94),
('#trumpvlog', 78),
('#trumpforpresident', 76),
('#draintheswamp', 75),
('#votetrump', 69),
('#covid19', 66),
('#trump2016https', 66),
('#kag2020', 63),
('#bigleaguetruth', 54),
('#2a', 50),
('#imwithyou', 48),
('#debate', 46),
('#2016', 46),
('#demdebate', 42),
('#gopdebate', 40),
('#usmca', 40),
('#trumptrain', 40),
('#crookedhillary', 38),
('#fitn', 38),
('#coronavirus', 36),
('#tbt', 35),
('#maga🇺🇸', 31)]
plot_cloud(list_top_hashtags)
Among the most used hashtags we find many important slogans used by Trump in the last decade: #trump2016 certainly related to the election, #makeamericagreatagain (and the acronim #maga) related to the 2016 election campaign and many other hashtags related to the world of politics.
We do an analysis of the most tagged people in Trump's tweets
rdd_at = rdd_trump_text_words.filter(lambda line : line.startswith('@'))\
.map(lambda word: (word,1))\
.reduceByKey(lambda x,y : x+y)\
.sortBy(lambda tuple: tuple[1], ascending=False)
list_top_at = rdd_at.take(30)
list_top_at
[('@realdonaldtrump', 10763),
('@foxnews', 927),
('@whitehouse', 836),
('@foxandfriends', 703),
('@barackobama', 567),
('@cnn', 393),
('@apprenticenbc', 390),
('@seanhannity', 309),
('@ivankatrump', 302),
('@teamtrump', 295),
('@mittromney', 268),
('@gop', 243),
('@donaldjtrumpjr', 221),
('@gopchairwoman', 218),
('@danscavino', 213),
('@nytimes', 206),
('@erictrump', 200),
('@breitbartnews', 180),
('@potus', 170),
('@jim_jordan', 166),
('@megynkelly', 156),
('@flotus', 151),
('@dbongino', 143),
('@marklevinshow', 141),
('@greggjarrett', 140),
('@oreillyfactor', 136),
('@loudobbs', 135),
('@celebapprentice', 128),
('@macys', 128),
("@barackobama's", 127)]
plot_cloud(list_top_at)
The most quoted person in Trump's tweets is himself (thanks to other people's retweets) and then mainly TV channels, newspapers or other people related to politics stand out.
We want to do an analysis of the years and hours of the day when Trump tweeted most: the problem is that some tweets are split over several lines, but since we are only interested in the dates, we filter out the lines where we only have the dates, and then extract them and delete the part of the text that comes before.
rdd_trump_tweets_cleaned = rdd_trump_tweets.filter(lambda tweet: ';' in tweet)
rdd_trump_tweets_cleaned.take(20)
['Republicans and Democrats have both created our economic problems.;02/08/2011 18:07:48', 'I was thrilled to be back in the Great city of Charlotte, North Carolina with thousands of hardworking American Patriots who love our Country, cherish our values, respect our laws, and always put AMERICA FIRST! Thank you for a wonderful evening!! #KAG2020 https://t.co/dNJZfRsl9y;03/03/2020 01:34:50', 'RT @CBS_Herridge: READ: Letter to surveillance court obtained by CBS News questions where there will be further disciplinary action and cho…;17/01/2020 03:22:47', 'The Unsolicited Mail In Ballot Scam is a major threat to our Democracy, & the Democrats know it. Almost all recent elections using this system, even though much smaller & with far fewer Ballots to count, have ended up being a disaster. Large numbers of missing Ballots & Fraud!;12/09/2020 20:10:58', "RT @MZHemingway: Very friendly telling of events here about Comey's apparent leaking to compliant media. If you read those articles and tho…;17/01/2020 13:13:59", 'RT @WhiteHouse: President @realDonaldTrump announced historic steps to protect the Constitutional right to pray in public schools! https://…;17/01/2020 00:11:56', '“I’m running as a proud Democrat, for the Senate”. Sleepy Joe Biden today. It’s only going to get worse. It is not sustainable for our County - China will own us!!!!;12/10/2020 22:22:39', 'Getting a little exercise this morning! https://t.co/fyAAcbhbgk;01/02/2020 16:14:02', 'https://t.co/4qwCKQOiOw;23/10/2020 04:52:14', 'https://t.co/VlEu8yyovv;23/10/2020 04:46:53', 'https://t.co/z5CRqHO8vg;23/10/2020 04:46:49', 'https://t.co/TQCQiDrVOB;23/10/2020 04:46:45', 'RT @EricTrump: https://t.co/NcrNdOSfIV;23/10/2020 04:30:19', 'Unemploym…";08/03/2020 04:01:46', "RT @TomFitton: Vindman's behavior is a scandal. He should be removed from the @RealDonaldTrump White House ASAP to protect our foreign poli…;07/02/2020 17:35:20", 'Thank you Elise! https://t.co/Y4Hb0zf5jk;09/01/2020 12:24:31', 'As per your request, Joe... https://t.co/78mzcfLEsF https://t.co/AalVkSfHFi;23/10/2020 04:09:59', "RT @marklevinshow: President Trump was outstanding.\xa0 I couldn't be prouder.\xa0 Biden was making faces and was very dishonest about some of hi…;23/10/2020 03:58:38", 'RT @RealRLimbaugh: Winning, winning, winning! @realDonaldTrump;23/10/2020 03:57:45', 'HUGE win today for the United States and for peace in the world. Sudan has agreed to a peace and normalization agreement with Israel! With the United Arab Emirates and Bahrain, that’s THREE Arab\xa0countries to have done so in only a matter of weeks. More will follow! https://t.co/UHB8H6oaZc;23/10/2020 16:55:29']
The number of dates present, also indicates the total number of tweets Trump has made from 2009 to 2020
rdd_numbers_of_tweets = rdd_trump_tweets_cleaned.count()
print('The data contain',rdd_numbers_of_tweets,'tweets')
The data contain 55514 tweets
We identify the month and year in which Trump tweeted the most
rdd_trump_date = rdd_trump_tweets_cleaned.map(lambda tweet: (tweet.split(';')[1]).split(' ')[0])\
.map(lambda line: line[3:])\
.map(lambda date : (date, 1))\
.reduceByKey(lambda a,b : a+b)\
.map(lambda a: (a[1], a[0]))\
.sortByKey(False)\
rdd_trump_date.take(20)
[(1415, '10/2020'), (1363, '09/2020'), (1126, '12/2019'), (1121, '08/2020'), (1121, '05/2020'), (1119, '06/2020'), (1119, '01/2015'), (1018, '10/2019'), (970, '03/2020'), (964, '01/2020'), (902, '04/2013'), (899, '02/2020'), (893, '02/2013'), (869, '07/2020'), (865, '07/2013'), (862, '11/2019'), (840, '04/2020'), (826, '03/2013'), (798, '04/2015'), (797, '09/2019')]
The period in which Trump tweeted the most was the autumn of 2020: although he was already president at that time, the election was imminent and Trump wanted to make his voice heard by the American people on Twitter as well.
We visualize the hours of the day in which Trump tweeted the most
rdd_trump_hour = rdd_trump_tweets_cleaned.map(lambda tweet: (tweet.split(';')[1]).split(' ')[1].split(':')[0])\
.map(lambda hour : (hour, 1))\
.reduceByKey(lambda a,b : a+b)\
.map(lambda a: (a[1], a[0]))\
.sortByKey(False)\
list_trump_hour = rdd_trump_hour.take(24)
list_trump_hour
[(3851, '12'), (3637, '20'), (3425, '19'), (3347, '13'), (3141, '11'), (3038, '14'), (2955, '18'), (2944, '02'), (2891, '21'), (2840, '15'), (2732, '01'), (2575, '03'), (2571, '16'), (2367, '22'), (2305, '00'), (2270, '17'), (2203, '23'), (1887, '04'), (1694, '10'), (890, '05'), (667, '09'), (482, '06'), (436, '08'), (366, '07')]
def plot_amount_tweet_per_hour(trump_hour):
pd_df = pd.DataFrame({
'Hour': [ t[1] for t in trump_hour ],
'Nb_of_tweet': [ t[0] for t in trump_hour ]
}).sort_values('Hour', ascending=True)
plt.figure( figsize = (15, 8) )
chart=sb.barplot(data=pd_df, x='Hour', y='Nb_of_tweet', order=pd_df['Hour'], color='red')
chart.set_xticklabels(
chart.get_xticklabels(),
rotation=80,
horizontalalignment='right',
fontweight='light',
fontsize='x-large',
)
plt.show()
plot_amount_tweet_per_hour(list_trump_hour)
Trump prefers to tweet in the early afternoon or in the evening: surprisingly, there is a fairly high number of tweets even at night, while he tweets very little in the morning.
We visualize the years in which Trump tweeted the most
rdd_trump_year = rdd_trump_tweets_cleaned.map(lambda tweet: (tweet.split(';')[1]))\
.map(lambda tweet: (tweet.split(' ')[0]))\
.map(lambda tweet: (tweet.split('/')[2]))\
.map(lambda year : (year, 1))\
.reduceByKey(lambda a,b : a+b)\
.map(lambda a: (a[1], a[0]))\
.sortByKey(False)\
rdd_trump_year.take(12)
list_trump_year = rdd_trump_year.take(12)
list_trump_year
[(11335, '2020'), (8144, '2013'), (7818, '2019'), (7536, '2015'), (5784, '2014'), (4225, '2016'), (3568, '2018'), (3531, '2012'), (2602, '2017'), (773, '2011'), (142, '2010'), (56, '2009')]
def plot_amount_tweet_per_year(trump_year):
pd_df = pd.DataFrame({
'Year': [ t[1] for t in trump_year ],
'Nb_of_tweet': [ t[0] for t in trump_year ]
}).sort_values('Year', ascending=True)
plt.figure( figsize = (15, 8) )
chart=sb.barplot(data=pd_df, x='Year', y='Nb_of_tweet', order=pd_df['Year'], color='purple')
chart.set_xticklabels(
chart.get_xticklabels(),
rotation=80,
horizontalalignment='right',
fontweight='light',
fontsize='x-large',
)
plt.show()
plot_amount_tweet_per_year(list_trump_year)
The years in which Trump tweeted the most were the years before the election, whereas he tweeted very little in the first years he was in office (2009, 2010) when social media were not yet mass media, and he tweeted very little even in the two years after he was elected president (2017 and 2018)