Political Sentiment Analysis on Twitter

Data:Case study Modi and Rahul Gandhi

Dharminder Yadav1, Dr. Umesh Chandra2
1
Research Scholar, Computer Science Department, Glocal University, Saharanpur, UP, India
2
Assistant Professor, Computer Science, Banda University Agriculture and Technology,
Banda, UP, India
1
[email protected]@gmail.com
their opinion and ideas on these websites daily.Facebook,
Abstract— Textual information is growing in the world Twitter, Whatsapp and other long-range informal
due to the use of social networking sites: Twitter, communication destinations today go about as a pool of
Facebook, Whatsapp, Web 2.0, telegram, web article and assessment since individuals are sharing their view on a
subject of intrigue. These assessments and thoughts are as
micro-blogging site. Today Twitter, telegram and
content, conclusion mining, content mining and feeling
Whatsapp are the best micro blogging sites. People were examination is use content information and grouped the
spending most of the time on social networking sites for slant based on content information as positive, negative
personal communication, chat and opinion sharing. They and impartial. Feeling examination and supposition
share the opinions about political persons, political party, mining is or was utilized in various associations, for
products, services and political events between users example, governments, social insurance, training,
using social networking sites. Opinion mining and building, and innovation segment, assembling, and retail
division to get the sentiments on their item and
sentiment analysis used text observations to know what
administrations. Content mining and conclusion mining
people are thinking about the products or political parties utilized Natural Language Processing (NLP) and Natural
etc. Social networking site produces a huge amount of Language Toolkit (NLTK) to mine the feelings on
unstructured text daily. Processing and extraction of specific themes, item Prediction of the decision, stocks
information from dynamically unstructured text, we need costs or administrations [1]. Slant investigation and
techniques such as text mining, machine learning, natural conclusion digging are fundamental for business the
language processing (NLP), information retrieval (IR) executives and for settling on the best choice based on
clients see/feeling removed from content information.
and Data mining. Twitter is the best source of an
Sentiment mining separate is arranged into three unique
unstructured source ? and used for sentiment analysis dimensions as Document level, Sentence level, and
and opinion mining. Aspect level. In report level examination, the entire record
estimation expresses as positive or negative. Commented [IR4]: Relate it with Rahul and Modi? You are
writing only theory ?
The main aim this paper is to analyse the people opinion
regarding Sh. Narender Modi from Bhartiya Janta Party In sentence-level sentiment analyses is subjective
(BJP) and Rahul Gandhi from (Indian National Congress classification and find the polarity of short sentences.
Entity/ Aspect level sentiment analysis is also called
(INC) by using the twitter tweets in the English language.
augmented analysis and find the sentiment of entity or
We used twitter API using R to get the tweets. R is a very aspect. Eg. “My laptop features are good but its Commented [IR1]: Some where past and somewhere present
powerful language for data analysis, data mining and tense?
processing is slow and keyboard is not user-friendly".
most probably used for sentiment analysis/ opinion “Laptop features are good” show positive sentiment but
mining. We are using corpus base and dictionary-based “processing is slow” show negative sentiment. Commented [IR2]: Present tense , check every where
method to find the sentiment of tweets. We try to show the Twitter is the most popular social networking site that
each and every sentiment of Twitter users towards Prime provides the real-time micro-blogging services to uses to
Minister Sh. Narender Modi from BJP (Bhartiya Janta share opinion/views/information in the form of text; this
Party) and Sh. Rahul Gandhi president of INC (Indian text is called the Tweets and maximum size of tweets are
National Congress) and classified the sentiment as 140 characters. Twitter was launch in 2006, but users
positive negative and neutral. sharing information and daily activity start in 2007[2]. At
a time Twitter’s has more than 467 million users and daily
share or generate more than 175 million tweets [3].
Keywords— Twitter,Sentiment Analysis,Election,
Tweets are collected in English language using Twitter
opinion mining, BJP, INC,social media,natural API that considered leader of two major parties of India
language processing, R, Twitter API. such as Sh. Narendra Modi from BJP and Rahul Gandhi
from INC and labelled the sentiment as a positive
I. INTRODUCTION negative and neutral by using the free tool R and its
libraries. Sh. Narendra Modi is dynamic and hardworking
Due to the advancementgeneration and use of social
Prime minister of India and Sh. Rahul Gandhi is young
networking website huge amount of structured, semi-
president of Indian National Congress. In the era of Prime
structured and unstructured information is produced every
Minister Modi BJP win 80% of election and BJP reach on
day and to be had on internet site and apps. User’s sharing Commented [IR3]: Write it properly or cut it
the peak. Modi provides the special status to India in the method to show the polarity as negative, positive and
world and 6th largest economy of the world. But in the era neutral. They uses Crawler-Gathers, Semantic
president of INC Sh. Rahul Gandhi wins three states engine,Search engine Machine translation engine, Geo-
election (name of states here).
referential engine, Classification engine, clusters and sub-
In sentiment analysis may be very hard to predict the end clusters, highlighting meaningful information.
result of the election at the opinion of customers. In this
paper we strive to answer the query using sentiment YasufumiTakama [6] in their paper proposed method to
analysis popularity of any political party on Twitter can predict the user’s TV watching behaviour using sentiment
growth the chance of winning the election? To reply to analysis. The technique creates positive and negative
this question we examine the twitter tweets and study the profile with bookmark design for every client dependent
people's sentiment in terms of positive, negative and on IEPG and evaluated rating of staring at the TV
neutral polarity. The authors collect the tweets, clear out program. On the off chance that client watches program
the tweets, pre-processing the tweets after which applying for long time that implies he/she is intrigued and if
the sentiment mining on tweets to expect the political changes channel as often as possible that implies he/she
state of affairs of these two leaders in India. isn't intrigued. Daniel e.o’ leary[7] in his paper focused on
blog mining-evaluations. He termed blog as type of media
In this paper, we will discuss the introduction, that lets in consumer to present a variety of statistics
sentiment analysis on twitter tweets, writing survey consisting of evaluations and personal views. Companies
political and general opinion investigation, the technique are searching upon blogs as an important source of
utilized for tweets extraction, pre-handling, and records and understanding. Blogs can also contain
assumption examination of tweets. In the result area, redundant and co related facts about any topic monetary,
examine the outcome and ordered the extremity in political, amusement and information.it represents a
graphical reorientation. This paper is finished up by single person or a collection. Weblog mining is a system
different difficulties present in notion examination of looking and studying blogs a good way to generate
utilizing Machine learning strategy and free accessible extra insights that would otherwise now not be found with
devices such as R, Python utilizing twitter and talking the aid of inspecting a single blog.Turney[8]and Pang et
about the future extent of opinion investigation. al.[9] started his research in the area of sentiment analysis
with product and movie reviews. Turney [8] used bag-of-
words method and Point wise Mutual Information (PMI)
II. LITERATURE REVIEW to estimate the sentiment orientation of phrases.
This sectionof paper explain the related study of opinion Po-Wei Liang et.al. [10] Used Twitter API to gather
mining and sentiment analysis in different languages, tweets from twitter and sifted through the tweets opinion.
related tools and technology, data collection and Unigram Naive Bayes demonstrate was produced for
processing methods, micro-blogging sites and algorithm extremity recognizable proof. They kill the undesirable
that required for these process.Researchers divideds the highlights by utilizing the Mutual Information and Chi
related work into two groups namely General sentiment square component extraction strategy. But the predicting
analysis research and political sentiment analysis the tweets as positive or negative did not give better result
research. This paper analysis the political sentiment using by this method.N Rochmawati and S C Wibawa[11]uses
English tweets to predict the users opinion about Prime twitter to know the conclusion of individuals about
Minister Sh. Narendra Modi and Sh. Rahul Gandhi Rohingya and supposition was ordered into 3 gatherings:
president of INC. positive, negative and nonpartisan. Gathering information
was utilized innocent Bayes, twitter utilizes R and its
Ayeena Malik et.al. [4]Experiment describes the
library to anticipate the conclusion of individuals and plot
sentiment analysis the people’s opinion when Prime
the outcome in chart which speak to all conceivable
Minister Narendra Modi visits to United States of
supposition of tweets. At last, there isn't clear sign of
America They collect the tweets from the twitter in .csv
tweets on the grounds that the vast majority of the tweets
format. Predict the tweets polarity as positive, negative
conclusion is obscure, positive versus negative tweets
and neutral.FedricoNeri [5] describes a sentiment analysis
equivalent and less unbiased
study performed on about 1000 Face book posts about
new casts and comparing the sentiments of Rai. They Eman M.G.Younis [12] utilizes twitter API and TwitteR
used supervised machine learning and unsupervised bundle to look through the twitter messages. They
machine learning sentiment analysis method. In exhibited an open source approach, all through which,
supervised machine learning manually assigned the twitter Micro websites information has been gathered,
polarity for training data set of text. But in unsupervised pre-prepared, broke down and envisioned utilizing open
machine learning uses a lexicon based sentiment analysis source devices to perform content mining and assumption
examination. They collected client suppositions and
online surveys around two goliath retail locations in the V.METHODOLOGY AND FRAMEWORK
UK in particular Tesco and Asda stores over Christmas
period 2014. The slant investigation of the client Figure 1 show the steps involved in mining the sentiment
sentiments makes it simpler for organizations to of twitter tweets. The steps describe in detail below:
comprehend their aggressive incentive in a changing
business sector and to comprehend their client sees about
their items and administrations, which likewise give an
understanding into future promoting techniques and basic
leadership strategies.

El-Kourdiet al. [13] used Naïve Bayes algorithm for

categorizing Arabic text documents. They categorized the
document into five pre-defined categories by using Naïve
Bayes categorizer.Tumasjan et. al. [14] centred on the Commented [IR5]: dot
2009 German federal election the use of twitter tweet and
predict election results and concluded that the variety of
tweets of a party is directly proportional to the chance of
prevailing the elections. They gathered over a hundred
thousand tweets between august and September months of
2009 containing the names of the six parties.Choy et al.
[15] uses online sentiment to predict the vote percentage
for each of the candidates in the Singapore presidential
election of 2011.Wang et al. [16] proposed a real-time
sentiment analysis system using Naive Bayes model with
unigram features for political tweets and collected the
sentiment annotations using Amazon Mechanical Turk
which was based on the U.S. presidential election of
2012. Mishra at.el[17] uses Machine learning,NLPand
Dictionary Bases approach method for sentiment analysis
of digital India. Tweets are collected from twitter using Fig 1: Sentiment analysis Diagram
Twitter API and classified the tweet as positive negative
and neutral. Result clearly indicates that the 50% tweet is A. Collecting Data
positive, 20% negative and 30% have neutral polarity.
To collect the twitter tweets we are using TwitteR API.
Some research has been done onlocal Indian languages We create the App account on twitter (Figure 2) using
such Hindi, Marathi, Urdu etc. for sentiment analysis. Das Twitter account to get twitter key and token which are
and Bandopadhya [18] were prepared a Bengali used by TwitteR libraryin R and return the tweets in text
format. Searched and retrieved tweets, from twitter
SentiWordnet and define the four procedures to
account was saved in database in .csv file format or excel
understand the sentiment of a word. Joshi et al. [19] format. We are used searchtwitter function to search the
created a lexical resource of Hindi SentiWordnet to tweets.
understand the sentiment of Hindi word and using the
methodologies such as Machine Translation, Resource
Based Assumption Analysis and Language Sentiment
Analysis with accuracy of 78.14%.Gune et al. [20] build a
parser in Marathi language. Mittal et al. [21] build a Hindi
language corpus, improve the present Hindi SentiWordnet
and generated efficient approach to identify the sentiment
from Hindi content with 80% accuracy. Ambati et.al.
[22], try to find the sentiment of Hindi data and get 76%
rate of error. Arora et. al. [23], describe the diagram based
system and used WordNet to collect the Hindi based Fig 2: App Account on twitter
subjective dictionary.
We used hashtagsto search the tweets related to Prime
minister Sh Modi and president of INC Sh Rahul Gandhi,
Hashtag are #modi, #narendramodi, #Rahulgandhi,
#rahuland # presidentINCto search the trending tweetson
Twitter which representing the political views of people.
We collected 10,000 tweets related PM Modi, 10,000
tweets from Rahul Gandhi.

B. Data Pre-processing

Datapre-processingsteps are: Steaming,

Tokenization,Stop Word Removal and Normalization. Fig 3: Tweet Pre-processing Steps
Wefirst remove irrelevant tweets and twitter features
from twitter data.Figure 3 show the twitter features,
special character and symbol such as, URL, Mention,
Recency, Hashtag, Emoticons, Friends, Followers and
Retweet.

Fig 4:Tweets after Cleaning

Sentiment Analysis: In this stage we perform

Fig 3: Twitter Tweets without pre-processing sentiment analysis on cleaned data and clarified the
emotion as a positive, negative and neutral using library
In steaming we identify and remove the suffixes and sentiment and tidytext in R. Figure 5 represent the
prefixes from each world of tweets like “lly”, ”tion”, emotions of each tweet with ten sentiment. Each and
”sub”, ”re”, “ion” etc. to identify the actual emotion of every tweet has ten emotions, zero show that word
tweets. Tokenization is the process which split thesystem contain in the tweet has no emotion but one or higher
of text into tokens. Tokens are the small words which value show that tweet has emotion out of ten emotion. We
have predefine meaning. White spaces and punctuation calculate the value of sentiment of each tweet, if value of
are used as token delimiters.It included several steps like sentiment greater than zero then tweet is positive, if value
Emoticons replaced with original meaning, remove white of emotion less than zero then sentiment is negative
spaces, blank tweets, remove URL, remove html links, otherwise neutral sentiment.
Retweet Abbreviations like LOL, Plz, OMG etc. are
replace with original meaning Pragmatics handling such
as shittttttttt as shit, oooooooh as oh,hiiiiiiiiiiiii as hi etc.
Stop Words are those words which have no meaning in
context of Sentiment and opinion analysis. But they are
used to provide the meaning for sentence. In stop word
removal step we remove the unused stop words such as
prepositions(to, by, in, over, under, a, the, an, etc.) and
conjunctions (for, and, nor, but, or, yet, so).

All steps are performed using the library such as tidyverse

Tidytext, Stringr RCurl, TwitteR, httr and tm with R.
Library can be downloaded from (https://cran.r-
project.org). Figure 4, 5shows the tweets pre-processing
steps and tweets afterpre-processing respectively. After
pre-processing unused and meaningless tweets removed,
only those tweets are left for we will find the sentiment.
Fig 5: Emotion of Tweets
 VI. Results and Discussion
In this section we will discuss the result of experiment in Sr. Emotions Number of Percentage
detail. First aim of this experiment is accuracy of No Tweets of Tweet
sentiment analysis. For this purpose we are using R and
1 Anger 2197 4.88%
its tools for processing of tweets and sentiment analysis
with wordcloud. Wordcloud is also called tag cloud/text 2 Anticipation 4234 9.4%
cloud and display the text data visually. These graph and 3 Disgust 3034 6.74%
Wordcloud display the textual data in informative and 4 Fear 2637 5.86%
interesting form.First we discuss the tweet sentiment of
INC President Sh Rahul Gandhi, after that we discuss the 5 Joy 3501 7.78%
tweets sentiment of Our Prime Minister Sh. Narendra 6 Sadness 4066 9.03%
Modi from BJP. The result of experiment and testing has 7 Surprise 1342 3%
been done is as follows: 8 Trust 7778 17.28%
9 Negative 7945 17.65%
10 Positive 8266 18.36%

Table 1:Ten Sentiment of Sh. Rahul Gandhi

Fig 6:Positive and Negative Tweet of Sh. Rahul

Gandhi

In Figure 10 Tweet no 7 is positive tweets in which

“presedent Mrs. Rahul Gandhi announced the manifeso”,
but tweet number 12 is negative tweet in which some one
say that “minimun income gurentee is like a dole”. Figure
7 show all the ten sentiment of Rahulgandhi President of
INC.All the sentiment are: anger,positive, negative,
disgust, anitcipation, fear, joy, trust,sadness, surprise.
Largest group of people with positive sentiment. Second
and third higest group are negative and trust sentiment
respectively. Table Number 1 show all sentiment in detail
with percentage. 18.36 percent of people with positive
sentiment and 17.65 % people with negative sentiment in
case of Sh Rahul Gandhi.There is minor difference
between positive and negative voter. But there is 17.28
percent tweets with trust sentiment, other seven
sentiment less than 10 percent.Its mean that Sh Rahul
Gandhi get sympathi and trust from tweeter user.

Fig 8: Sentiment of Sh. Rahul Gandhi

Figure 8 show the sentiment of tweets #rahulgandhi. We

found 14980 tweets, in which 4708 positive,
2204negative and 8088 neutral tweets. Inour result most
of the tweets are neutral; these neutral tweets are the
combination of rest eight sentiment of tweet.Figure 9
show the wordcloud of #rahulgandhi.Wordcloud is the
graphical representation of text. Wordcloud is a package
available in R which provides the visual representation of
text data. Wordcloud is also known as tag cloud or text
cloud. In this wordcloud show all the sentiment of tweets
and related word with sentiment in different colour.
Fig 7: Ten Sentiment of Sh. Rahul Gandhi
 Sr. Emotions Number Percentage
No of of Tweet
Tweets
1 Anger 1642 6.77%
2 Anticipation 3215 13.26%
3 Disgust 748 3.08%
4 Fear 1446 5.96%
5 Joy 2231 9.20%
6 Sadness 1467 6.05%
7 Surprise 1407 5.80%
8 Trust 3484 14.37%
9 Negative 3070 12.67%
10 Positive 5519 22.77%

Table 2:Ten Sentiment of Sh. Narendra Modi

Fig 9: Wordcloud of #rahulgandhi
Above Table 2 show all the sentiment of tweets with
percentage and value of each sentiment. Figures 12 show
the positive, negative and neutral opinion group. Negative
tweets are half of the positive tweets, but most of the
tweets are in neutral group.Figure13 show wordcloud of
Prime Minister Modi.

Fig 10: Positive and Negative Tweet of Sh.

Narendra Modi

Figure 10; show the positive and negative tweet of collect

ed tweets. The negative tweet is "A person does not alway
s get depressed all of a sudden there are signs that show a
person is heading into depression” and the positive tweet
is “Prime Minister NarendraModi shares his success Man
tra with the young students and he revealed a few nameso
f his class" .We collected 15000 tweets related to Prime m
inister Narendra Modi. Figure 11; show all the ten sentim
ent of tweets related to modi. Most of the peoples thinkin
g positive and trust regarding modi, but some are negative
.

Fig 12: Sentiment of #narendramodi

Fig 11: Ten Sentiment of #narendramodi Fig 13: Wordcloud #narendramodi

 VII. CONCLUSIONS ACM International Conferences on Web Intelligence
In this is a technical paper we used twitter to collects and Agent Technology, 2007.
tweets, to perform sentiment analysis. We constructed [7] Daniel, E., Blog mining-review and extensions.
corpus for ten sentiments of tweets,anger,positive, Decision Support Systems, Elsevier, 2011.
negative, disgust, anitcipation, fear, joy, trust,sadness,
surprise and then concluded these ten sentiment into three
sentiments such as positive, negative, neutral tweets and [8] Peter D. Turney. Thumbs up or thumbs down?:
create the wordcloud. We use the R language and Rstudio semantic orientation applied to unsupervised
to implement sentiment analysis. It is open tool for text classification of reviews, . 2002.
mining and sentiment analysis for twitter data to analysis
user contribution reviews for product and services. It also [9] Pang B, Lillian Lee, and Vaithyanathan S.K.,
helps to improve the business value and customer Thumbs up?Sentiment classification using machine
learning techniques. In Proceedings of the
relationship. Based on the result get from above
conference on Empirical Methods in Natural
experiment, we conclude that most of the tweets are
Language Processing – Vol. 10, 2002.
neutral few tweets can be positive and negative. In this
experiment Prime Minister Modi get 22.77% of positive [10] Po-Wei Liang, and Bi-Ru Dai, “Opinion Mining on
and 12.67% of negative tweets. Its mean twitter user Social Media Data", IEEE 14th International
thinking was positive regarding modi. Users trusted on Conference on Mobile Data Management, Milan,
Modi because trust sentiment is 14.37%. But in case of Italy, June 3- 6, pp. 91-96, 2013.
INC president Sh Rahul Gandhi difference between
positive and negative tweets is less than 1%. Its mean [11] N. Rochmawati and S.C. Wibawa, Opinionanalysis
people confused to vote for Rahul Gandhi or not. From on Rohingya using Twitter data, IOP
the above result it is very difficult to judge who will win Conf.Series:Material Science and Engineering, 2013.
the election 2019.Our motive to analyse the Twitter users
thinking only. In sentiment analysis some technical issue [12] Eman M.G.Younis,Sentiment Analysis and Text
are present due toincompleteness of dictionary or other Mining for Social Media Microblogs using Open
reasons. We additionally analyse that there's a want to Source Tools: An Empirical Study,International
enhance twitter sentiment analysis and text mining. With Journal of Computer Applications (0975 – 8887)
these technical trouble, huge range of industries are used Vol. 112 – No. 5, Feb. 2015.
social media-web sites for opinion mining and sentiment
analysis. [13] El-Kourdi, M., Bensaid, A., and Rachidi, T.,
Automatic Arabic Document Categorization Based
References on the Naïve Bayes Algorithm, In proceedings of
COLING 20th Workshop on Computational
[1] Berry Michael W., Automatic Discovery of similar Approaches to Arabic Scriptbased Languages,
words, in Survey of text mining clustering, University of Geneva, Geneva, Switzerland, August
classification and Retrieval, Springer Verlag, New 23rd— 27th., 2013.
York, LLC, pp. 24-43, 2004.
[14] Tumasjan, A., Sprenger, T. O., Sandner, P. G.,
[2] Malik1 A., Kapoor D.,and SinghA. P., Sentiment andWelpe, I. M. Predicting Elections with Twitter:
Analysis on Political Tweets, International What 140 Characters Reveal about Political
Symposium on “Fusion of Science & Technology”, Sentiment, ICWSM, 179-185, 2010.
New Delhi, India, pp.359-361,2016.
[3] BoydD., GolderS., and Lotan,G., Tweet, tweet, [15] Choy, M., Cheong, L. F. M., Ma, N. L., and Koo, P.
retweet: Conversational aspects of retweeting on S. A sentiment analysis of Singapore Presidential
twitter, System science (HICSS), 2010. Election 2011 using Twitter data with census
correction, 2013.
[4] Malik A., KapoorD.,and Singh A. P., Sentiment
Analysis on Political Tweets 5th, International [16] Wang, H., Can, D., Kazemzadeh, A., Bar, F., &
Symposium on “Fusion of Science & Technology”, Narayanan, S. A system for real-time twitter
New Delhi, India, January 18-22, 2016. sentiment analysis of 2012 us presidential election
[5] Neri, F., Aliprandi,C.,Capeci, F., and Caudro Tomas, cycle. In Proceedings of the ACL 2012 System
Sentiment Analysis on Social Media. ACM Demonstrations (pp. 115-120). Association for
International Conference on advances in Social Computational Linguistics, 2012.
Network Analysis and Mining, 2012.
[17] Mishra P., Rajnish R., and Kumar P, Sentiment
[6] Takama, Y., and Moto,Y., Profile Generation from Analysis of Twitter Data:Case Study on Digital
TV Watching Behavior using Sentiment Analysis. India, International conference on information
technology, pp.148-153, 2016.
[18] DasA, andBandyopadhyay S, “SentiWordNet for
Bangla,” Knowledge Sharing Event-4: Task, Vol. 2,
2010.

[19] JoshiA., B. A. R, and BhattacharyyaP., “A fall-back

strategy for sentiment analysis in Hindi: a case
study,” Proceedings of ICON 2010: 8th International
Conference on Natural Language Processing, Dec.
2010.

[20] GuneH., BapatM., KhapraM. M., and

BhattacharyyaP., "Verbs are where all the action lies:
Experiences of shallow parsing of a morphologically
rich language", Proceedings of the 23rd International
Conference on Computational Linguistics, pp. 347–
355, 2010.

[21] MittalN., Agarwal B., ChouhanG., BaniaN., and

PareekP., “Sentiment Analysis of Hindi Review
based on Negation and Discourse Relation,”
Proceedings of International Joint Conference on
Natural Language Processing, pp. 45–50, 2013.

[22] AmbatiB. R., HusainS., JainS., SharmaD. M., and

SangalR., “Two methods to incorporate local
morphosyntactic features in Hindi dependency
parsing,” Proceedings of the NAACL HLT 1st
Workshop on Statistical Parsing of
Morphologically-Rich Languages (SPMRL), pp. 22–
30, 2010.

[23] Arora,P. BakliwalA. and VarmaV., “Hindi

Subjective Lexicon Generation using WordNet
Graph Traversal,” International Journal of
Computational Linguistics and Applications, Vol. 3,
No. 1, pp. 25–39, 2012.

[24] https://cran.r-project.org

[25] https://www.rdocumentation.org

[26] https://www.rdocumentation.org/packages/wordclou
d/versions/2.6/topics/wordcloud.