IRJET- Real Time Sentiment Analysis of Political Twitter Data using Machine Learning Approach

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 4124
Real Time Sentiment Analysis of Political Twitter Data
Using Machine Learning Approach
Joylin Priya Pinto1, Vijaya Murari T.2
1Department of Computer Science and Engineering, NMAMIT, Nitte, Karnataka, India
2Professor, Department of Computer Science and Engineering, NMAMIT, Nitte, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Social media is an interesting platform to directly
measure people’s feelings. Communicationtechnologiesplaya
vital role in geographically locating these emotions. But, the
understanding is not a simple task. Generation of voluminous
data makes the manual process complicated. Social media
data face the problem of diversity of language; due to which
the automatic approach also becomes a tedious job. Social
network deals with controversial discussions on variety of
topics. These discussions help in the field of data analysis. In
this paper, we are going to analyze tweets on political
Ayodhya issue. Tweets of users are collected and analysis is
done with the help of machine learning algorithm, to classify
the polarity of tweets.
Key Words: Sentiment Analysis, Feature Extraction,
Support Vector Machine, Random Forest, Naïve Bayes,
Linear Regression, KNN
1.INTRODUCTION
Analyzing the sentiments of people is turning into a critical
viewpoint in a wide range of decision makingprocesssinceit
is useful in recognizing individual’s issues and strategies
strengths. Without a doubt, theseinformationcanbeutilized
to settle on increasingly educated choices which will
probably conclude in better utilization of assets, good
association, better administration,improvedcitizenlifestyle,
nice human relations and, in the long run, better society.
Taking are of large scale classificationissuesisveryessential
in numerous applications, for example, text classification
problem. There is a high possibility of facing distinctive
troubles while performing sentiment analysis; for each
situation people may not present their assumptions
similarly, a solitary sentence may be certain for one event
and can be negative for other event; an enormousnumber of
sentence mixes are possible. Identifying incorrect spelling,
and dealing with intensifiers and fake sentences are very
challenging.
In the past few years, people feedback and
sentiments were analyzed by opinion pooling system that
means with the help of conducting interviews,
questionnaires and by collecting opinions through forms,
[1][2] now a days social media is the best way to analyze
people sentiments [3]. There is an immense growth of user-
generated data in the form of blogs, forum and tweets. With
the increased usage of these platforms, people started to
speak about all matters: from personal to public; general to
specific matters. Social media is an effective platform to
understand individual sentiments.Theanalysisofuser posts
can be utilized to take proper decisions in a variety of fields
such as Business, Election, Productreview,Government,and
so on. Now a days, utilizing social media for political
discussions has become a common practice. Political
campaigns have misused immense range of information
accessible on Twitter to draw insights about people
sentiments and in this way structuretheirpromoting efforts.
Different sentiment analysis algorithms can be utilized to
distinguish and break down the attitudes of the users
towards a political talk. Furthermore, with the recent
advancements in machine learning algorithms, it became
possible to enhance the exactness of sentiment analysis
predictions.
An effective online social micro-blogging service is
Twitter. It allows users to communicatewithshortmessages
of length 140-characters. These messages are called as
"tweets". Twitter is speciallyaninteresting platform because
of the usage of its hash tags. Along with the short messages,
users can use hash tag symbol ‘#’ before a specific keyword
or phrase in the Tweet to arrange the tweets and makethem
easier in Twitter Search. The issue of text classification can
be made relatively simpler since the hash tag itself can
express a feeling or sentiment. In this work, twitter data on
Ayodhya issue is used as data source for sentiment analysis.
1.1 Sentiment Classification
Sentiment classification is one of the important topics in
sentiment analysis, which classifies the expressed opinion
towards an entity as positive, negative or unbiased. Three
most essential categorical levels in Sentiment Analysis are
Document level, Sentence level, and Feature based Analysis
[4]. The underlying stage in Sentiment Analysis istoperceive
whether the sentence is opinionated or objective. Subjective
sentence conveys emotions whereas objectivesentencedoes
not convey any emotion; usually considered as unbiased and
also being ignored in the analysis. Sentence level
classification is different from that at document level [5]. A
document can be more or less opinionated, whereas a
sentence can be only subjective or objective.Comparingwith
the document level sentiment classification, sentence level
classification has one more task to do. There is a need to

process the sentences containing no opinion with sentences
containing opinions before classifying on the objects and
their features to positive or negative.
1.2 Entity Discovery and Extraction
People communicate their opinions and sentiments against
an entity or object. Sentiment analysis and opinion mining
can be formalized as: “Given a set of evaluative text
documents D that contain sentiments about an object,
sentiment analysis and opinion mining aim to extract
attributes and components of the object that have been
commented on in each document d in D and to determine
whether the comments are positive, negative or neutral.” To
perform sentiment analysis, along with distinguishing
between polarities of opinions,identifyingcorrelatedentities
with which opinions are expressed is also very important.
Because without analyzing about which entity each sentence
talks about, the mined opinion is meaningless.
2. LITERATURE REVIEW
Many researchers have done lot of work on “Sentiment
analysis“ in past few years. Since the starting of the century
itself the work in the field hadbegun. Inthebeginningperiod,
sentiment analysis was intended for binary classification,
where opinions were assigned to bipolar classes as positive
or negative.
Extensive consideration has been given as of now to
the exploitation of big datasets generated by the users
through social media platforms withwhichsocialbehaviorof
the people can be identified. In order to naturally recognize
people sentiments, few approaches such as mentioned in [6]
use linguistic analysis systems; some depend on manual
approach [7] and some other depends on a self-loader
approach [8]. All the methods have pros and cons. The
manual approach provides more accurate result. But, it does
not work well while dealing with huge volumes of data; on
the other hand, an automatic approach is difficult to
implement. Because it has to deal with natural language
processing; there is a big challenge to work with the
characteristics of the posted content. For example, if tweets
are considered, they are normally posted with hashtags,
emoji’s and urllinks, whichmakesitdifficulttodeterminethe
expressed sentiment [9]. Moreover, training is needed to
automatic process. Training requireslargedatasetoflabelled
posts or lexical database where opinionated words are
labelled with sentiment values.Theseresourcesareavailable
for the English language [8], [9], [10], but the availability of
dictionary based dataset is very limited to work with other
languages [7], [11].
A. khan et.al [12] used an approach to compare between
positive and negative sentences. Itextractsinformationfrom
the Web and manually label the word set whichrequiresalot
of unnecessary effort. He has used a rule-based method for
sentiment analysis, which extract the overall document
polarity of specific words by a SentiWordNet dictionary, and
adjust itaccordingtothecontextinformation.First,sentences
are divided into subjective and objective ones on the lexical
dictionary basis. The subjective sentences are then further
processed to categorize as positive, negative or neutral
comments.
Judith et.al [13] focused on twitter dataanalysistechniques
to extract public opinion.Usingmachinelearningalgorithm,a
feature vector is constructed with the emotion describing
words from tweets and are fed to the classifier that classifies
the sentiment or opinion. It said that various twitter data
analysis techniquesare based on dictionaryandareusingthe
machine learning approaches.
Barbosa et. al. [14] designed a 2-step automatic sentiment
analysis method for classifying tweets. They used a noisy
training set to reduce the labelling effort in developing
classifiers. Firstly, they classified tweets into subjective and
objective tweets. Afterthat,subjectivetweetsareclassifiedas
positive and negative tweets.
Pak et. al. [15]created a twitter corpus byautomatically
collecting tweets using Twitter API and automatically
annotating those using emoticons. Using that corpus, they
built a sentiment classifier based on the multinomial Naive
Bayes classifier that uses N-gram and POS-tagsasfeatures.In
that method, there is a chance of error since emotions of
tweets in training setare labelled solelybasedonthepolarity
of emoticons. The training set is also less efficient since it
contains only tweets having emoticons.
Upma Kumari et. Al. [16] carried out a detailed study on
the sentimental analysis techniques and tools which have
been used in sentiment analysis process. She observed the
completeprocess that showshowtheinputisbeingclassified
in various stages of sentiment analysis. Accordingtoher,first
step of text summarization process is sentimental
information retrieval, then classification and finally
sentiment summarization process. Subjectivity of text is
checked first and based on the polarity, sentimentresultsare
generated. The proposed method providesimportantphases
of sentiment analysis on text classification.
Seyed-Ali et. al. [17] proposed a novel approach to
sentiment analysis of short informal texts such as tweets
posted in twitter. He used state-of-art Sentiment analysis
techniqueagainsta novel hybrid approach. The method uses
a sentiment lexicon to generate a set of new features to train
a SVM classifier. Author focused on the processing of twitter
data. Then tried to apply Support Vector Machine, Naïve
Bayes as well as Maximum Entropy classifiers. Accuracy of
the model iscompared against SVMandhybridapproachand
author proved that the novel hybrid method which he
proposed is more accurate than the support vector machine
model. Also the features presented in this approach does not
require more timeto compute the result which is better than
other techniques.
Devika et. al. [18] proposed her views on sentiment
analysis as it is a process of human feelings and opinion
extraction. It poses as the most powerful tool to get useful

information from the users which can be then used to
aggregate the collected sentiments from reviews. She has
discussed different techniques of sentiment analysis by the
analysis of various methods. Machine learning algorithms
such as SVM, N-gram, Naïve Bayes, KNN, Feature driven
sentiment analysis as well as rule and lexicon based
approaches are focused. The purpose of the paper was to
come out with a best technique to text classification
approach.
3. RELATED WORK
3.1 Conceptual Sentiment Analysis Framework
Twitter Sentiment Analysis Model does the analysis of
sentimentsas anautomatic process. It identifiestheopinions
as positive or negative and measures the strength of
expressed opinion on a topic or an entity. The conceptual
model of Twitter Sentiment Analysis has the following
modules:
 FeatureSelection Module which does theopinion
extraction job. It extract the words that express
opinions towards relevantentityatsentencelevel
analysis.
 Sentiment Detection Module which maps the
expressed opinion with respect to relevant entity
in each sentence.
 Sentiment Summation and Score Calculation
Module to measure the sentiment scores for each
entity.
Fig. 1 illustrates the conceptual framework of Twitter
Sentiment Analysis. Thedescription of each module is stated
below.
Fig -1:The Conceptual Sentiment Analysis Framework [20]
Data Collection: In the data collection process, data from an
intended source is collected in a systematic way. Then the
collected information is processed to produce better
outcome. Here, thissentiment analysis study is especiallyfor
political domainandforthepoliticalrelatedanalysis,political
based comments are required; political tweets can be tweets
related to discussion between political parties, aboutfamous
political leader, discussions on upcoming elections or any
trending political issues. Twitter allows to gatherdataforthe
academic or research purposes. Relevant data can be
collected by providing searchkeywords in the twitter search
field. In order to obtain twitter data, there is a need to create
Twitter developer account. Twitter requires the user to be
registered with the developer account and once the
registration is done, it provides necessary authentication
credentials when user queries the Twitter API. The user
account need to be verified and only if the account is valid,
twitter allows the user to access the API.
Here, the analysis is mainly on ‘Ayodhya issue’. The
tweets are scraped from the Twitter with the help of user
credentials. As there isnoavailabilityofpolaritybasedIndian
political training dataset, a new training dataset is built by
using Vader Sentiment Analyzer. Initially, training is done
based on this dataset for the collected twitter data. After
testing, the predicted tweetsare addedtothetrainingsetand
newly scraped tweets are kept for testing purpose.
Data Pre-processing Module: Text pre-processing is the
stage, where the raw data is transformed intosomethingthat
a classifier can read. User-generated data may contain URL
links, unnecessary noisy data. Pre-processing cleansthedata
by removing URL’s,usernames,stop-wordsandallirrelevant
data that does not express any sentiment. Here, the research
work uses hash tags and emoticons in the analysis of
sentiments in order to improve the efficiency.
In the sentiment analysis process, text pre-
processing plays vital role. No doubt, qualitative data
increases the efficiency where as an incomplete and noisy
data reduces the accuracy and effectiveness.
Feature Extraction: Most of the times sentiment analysis
relies on “Bag-of-Words” model, which is the representation
of text that describes the occurrence of words within a
sentence or document [19]. This is a very common feature
extraction method. Here, in this sentiment analysis model,
only the sentiment bearing wordsareextractedasfeaturesin
order to submit to the classification algorithm; remaining
words are ignored. Opinionated words usually express
subjective opinions. Here, in the feature extraction process,
words which exhibits a desirable state(e.g. good, wonderful)
express positive orientation whereas the words with
undesirable state shows a negative orientation. (e. g. very
sad)
Sentiment Analysis Method: Consider a set of tweets ‘T’,
which has a set of sentences S = {s1, s2, ... si} and each
sentence st express some opinions on an entity e. An entity
could be a person, a place, a topic etc. In this study, entity is a
topic i.e. “Ayodhya” issue.Eachsentences1maycontainsetof
opinionated words such as s1 = {w1, w2, ... wi}. Firstly,
sentiment score of each sentence will be calculated based on
expressed words. Secondly, sentiment summation will be

done to obtain the total sentiment scores towards entity e.
[20]
A sentence can express positive, negative or neutral
opinion. Here, sentiment score of each sentence is calculated
with the help of Vader Sentiment analyzer to distinguish the
sentences into positive or negative.Polarity_scores()method
is used to obtain sentiment polarity score. Sentimentscoreof
individual sentence will fall in the range of [+1, -1].
Compound score is a metric, which calculates the sum of all
lexicon ratings.
Table -1: Compound Score Metric
Sentiment Scoring
Positive Compound score>= 0.05
Neutral (Compound score >= -0.05) and
(Compound score < 0.05)
Negative Compound score <= -0.05
4. IMPLEMENTATION DETAILS
Machine Learning Approach
Machine Learning is a classification method, where the
objectsare represented by their features. In textanalysis,the
document is represented based on the words; i.e. words are
considered as features. Vectorization will translate the text
document into vector features. Different words are assigned
with proper feature weights and TF-IDF (Term Frequency –
Inverse Document Frequency) is the most common method
to allot proper weightage to the words.Testdataisconverted
into TF-IDF feature matrix. The number of feature
occurrences will be counted.Wordswithrareoccurrenceand
words with very high occurrences such as stop words which
are very frequent in a text document will be neglected and
they will be removed as they does not hold any sentiments.
Firstly, feature weights are extracted from the training
dataset. Then these features are applied on the test dataset
similar to training dataset.
Support Vector Machine: Support Vector Machine is a
promising new strategy for the arrangement of both linear
and non-linear classification. It is considered as the best
classification technique with respect to text classification. It
looks for the separating hyperplane. The maximum possible
margin between two classes can be considered as best
hyperplane. The hyperplane utilizes set of vectors. Support
vector machine is a supervised machine learning algorithm
which effectively handles text classification problems.
Sometimes, it is not possible to linearlyclassifyclasseswhich
are closer to one other. Therefore, non-linear classification
hyperplane will be used at this point to solve classification
problem. SVM is heavily used in analyzing sentiments;
Because of the sparse nature of text, it is ideally suitable for
SVM.
A. Dataset
For the implementation purpose, datahasbeenscrapedfrom
twitter with the help of usercredentials. Whilescrapingdata,
text or tweet is evaluated against Text Blob and Vader
sentiment analyzer. As there is no availability of readily
available political dataset, there was a necessity to create
training dataset manually. But, due to the need of large
dataset, it is not possible to manually label text with positive
and negative polarity. Therefore, data is scraped and
evaluated with natural language processing tools like Text
Bloband Vader sentiment analyzer. As persurvey[21],Vader
sentiment analyzer is more efficient natural language
processing tool. Therefore, Vader is considered here to
identify text polarity. Then the training dataset is created by
analyzing polarity scores and labelled with positive[1] and
negative[0] annotations.
B. Data Loading and Model selection
Training dataset is of 2631 tweets. Here, scikit learn tool is
used to import the dataset. Real time data is fetched and
tested against training data. Learning model is implemented
using Support Vector Machine algorithm. SVC( ) function of
sklearn python library is used to import the model. Test data
size is 800 tweets. Model predicted new test file with an
accuracy of 82%. Out of 800 tweets 760 tweets are predicted
as negative and 40 tweets arepredicted as positive. Textpre-
processing is still have to be improved. Graphical
representation of prediction gives clear cut idea on this text
classification.
Fig -2: Prediction result
Word Cloud representation of test data is plotted in order to
identify the importance of words. This representation helps
to identify highlighted words with no proper sentiment
values which can be ignored in the pre-processing stage.

Fig -3: Word Cloud Representation
5. CONCLUSIONS
Proposed system performs an effective sentiment
analysis on political reviews of population collected from
Twitter. Supervised machine learning technique is used for
classification purpose. Sentiment analysis with supervised
learning approach, mainly depends on the training set.
Ultimately, the problem of Sentiment Analysis istodeal with
natural language. It is very difficult to obtain patterns or
sentiments from a naturally written language.Inthepresent
world, subjective data is continuouslygrowing withoutlimit.
Hence, there is a need to keep track of this data which is an
extremely good source of information in the process of
decision making. Therefore, the proposed system is trying
implement a new approach for sentiment classification of
text.
In future, the plan is to come up with a technique
which enables privacy based sentiment analysis on social
media data such as handling political tweets. With the
adoption of this algorithm, the system should be able to
encrypt the too negative comments,whichareharmful tothe
society in terms of social or political public emotions.
Thereby, the proposedsystemhelpsinmaintainingharmony
in the society. Twitter is the largest social media network,
which generates millions of tweets daily on varietyofissues.
Therefore, in the future Hadoop framework whichworksfor
distributed Big data, can be used for handling political
twitter data.
REFERENCES
[1] Marco Furini, Manuela Montangero, “TSentiment: .On
Gamifying Twitter Sentiment Analysis”, IEEE ISCC 2016
Workshop: DENVECT, IEEE 2016, ISSN: 978-15090-0679-
3/16.
[2] G. Galster, “The mechanism(s) of neighbourhood effects:
Theory, evidenceandpolicyimplications,”inNeighbourhood
effects research: New perspectives. 2012, pp. 23–56.
[3] M. Montangero and M. Furini, “Trank: Ranking twitter
users according to specific top ics,” in Proceedings of the
12th Annual IEEE Consumer Communications and
Networking Conference (CCNC 2015), Jan 2015, pp. 767–
772.
[4] Walaa Medhat, Ahmed Hassan and Hoda Korashy,
“Sentiment analysis algorithms and applications: A survey”,
Shams Engineering Journal (2014) 5, 1093–1113.
[5] B. Liu. Handbook of Natural Language Processing,
chapter Sentiment Analysis and Subjectivity. Second edition
edition, 2010.
[6] M. Furini, “Users behavior in location-aware services:
Digital natives vs digital immigrants,” Advances in
HumanComputer Interaction, vol. 2014, 2014. [Online].
Available: http://dx.doi.org/10.1155/2011/678165
[7] L. Mitchell, M. R. Frank, K. D. Harris, P. S. Dodds, and C. M.
Danforth, “The geography of happiness: Connecting twitter
sentiment and expression, demographics, and objective
characteristics of place,” PLoS ONE, vol. 8, no. 5, 2013.
[8] C. Bosco, L. Allisio, V. Mussa, V. Patti, G. Ruffo, M.
Sanguinetti, and E. Sulis, “Detecting happiness in italian
tweets: Towards an evaluation dataset for sentiment
analysis in felicitta,” in Proc. of the 5th Workshop on
Emotion, Social, Signals, Sentiment &LinkedOpenData,May
2014.
[9] J. Carrillo de Albornoz, L. Plaza, and P. Gervas,
“Sentisense: ´ An easily scalable concept-based affective
lexicon for sentiment analysis,” in The 8th International
Conference on Language Resources and Evaluation (LREC
2012), 2012.
[10] P. Nakov, S. Rosenthal, Z. Kozareva, V. Stoyanov, A.
Ritter, and T. Wilson, “Semeval-2013 task 2: Sentiment
analysis in twitter,” in Proc. of the Workshop on Semantic
Evaluation. Association for Computational Linguistics, June
2013.
[11] Y. H. Hassan Saif, Miriam Fernandez and H. Alani,
“Evaluation datasets for twitter sentimentanalysis:Asurvey
and a new dataset, the sts-gold,” first ESSEM workshop,
2013.
[12] A. khan, B. Baharudin, ”Sentiment Classification Using
Sentence-level Semantic Orientation of Opinion Terms from
Blogs,” Processed on National Postgraduate Conference
(NPC), pp. 1 – 7, 2011.
[13] Judith Sherin Tilsha S., Shobha M S, “A Survey on
Twitter Data Analysis TechniquestoExtractPublicOpinion”,
International Journal of Advanced Research in Computer
Science and Software Engineering, Volume 5, Issue 11,
November 2015, pp 536-540.
[14] L. Barbosa and J. Feng, “Robust sentiment detection on
twitter from biased and noisy data,” in Proceedings of the
23rd International ConferenceonComputational Linguistics:
Posters, pp. 36–44, Association for Computational
Linguistics, 2010.
[15] A. Pak and P. Paroubek, “Twitter as a corpus for
sentiment analysis and opinion mining,” in Proceedings of
LREC, vol. 2010, 2010.
[16] Upma Kumari, Dinesh Soni, Dr. Arvind K Sharma, “A
Cognitive Study of Sentiment AnalysisTechniquesandTools:
A Survey,” in the International Journal of Computer Science
and Technology-Volume 8, ISSUE 1, JAN-MARCH 2017.

[17] Seyed-Ali Bahrainian, Andreas Dengel, “Sentiment
Analysis using Sentiment Features,” in the 2013
IEEE/WIC/ACM International Conferences on Web
Intelligence (WI) and Intelligent Agent Technology(IAT)
- 978-1-4799-2902-3/13 $31.00 © 2013 IEEE.
[18] Devika M D, Sunitha C, Amal Ganesha, “Sentiment
Analysis:A Comparative Study On DifferentApproaches,”
in the Fourth International Conference on Recent Trends
in Computer Science&Engineering.Chennai,Tamil Nadu,
India, published by Elsevier B.V.
[19] Deepu S, Pethuru Raj, S.Rajaraajeswari, “A
Framework for Text Analytics using the Bag of Words
(BoW) Model for Prediction”, in the 1st International
Conference on Innovations in Computing & Networking
(ICICN16), CSE, RRCE
[20] Xujuan Zhou, Xiaohui Tao, Jianming Yong,
“Sentiment Analysis on Tweets for Social Events”, in the
Proceedings of the 2013 IEEE 17th International
Conference on ComputerSupportedCooperativeWork in
Design, 978-1-4673-6085-2/13/$31.00 ©2013 IEEE.
[21] Hutto, C.J. & Gilbert, E.E. (2014). VADER: A
Parsimonious Rule-based Model for Sentiment Analysis
of Social Media Text. Eighth International Conference on
Weblogs and Social Media (ICWSM-14). Ann Arbor, MI,
June 2014.

IRJET- Real Time Sentiment Analysis of Political Twitter Data using Machine Learning Approach

More Related Content

What's hot (18)

Similar to IRJET- Real Time Sentiment Analysis of Political Twitter Data using Machine Learning Approach (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Real Time Sentiment Analysis of Political Twitter Data using Machine Learning Approach