Election Result Prediction using Twitter Analysis

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2357
Election Result Prediction using Twitter Analysis
Ajay Rao1, Varun Kanade2, Chinmay Motarwar3, Prof. Shital Girme4
1, 2, 3Undergraduate Student, Dept. Computer Engineering, Pune Institute of Computer Technology, Pune, India
4Professor, Dept. Computer Engineering, Pune Institute of Computer Technology, Pune, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Elections in India has always been considered
as an important event and has been keenly followed by
majority of people. The rapid increase of social media in the
recent past has provided end users a powerful platform to
voice their opinions. Twitter, being one such platform,
provides day-to-day updates on political events through
different hashtags and trends. People provide their opinion
by reacting on such political events. Our approach is to
gather a collection of tweets of top political parties
contesting within the General State election, 2022, then
compute the sentiment score. Dataset contains mixture of
both popular as well as recent tweets related to specific
political party. Specific keywords are used to extract tweets
for a party like ‘BJP elections 2022’, ‘#UPelections BJP’,
‘#Punjabelections BJP’, etc. We utilized a combination of
VADER Sentiment Analyzer and classic machine learning
algorithms like Random Forest Classifier, SVM, etc. to build
our classifier and classify the test data as positive, negative
and neutral tweets. Therefore, this work analyses tweets
collected from twitter and predicts election results by
performing sentimental analysis on them.
Key Words: Sentimental Analysis, Twitter, Supervised
learning, Natural Language Processing, Machine
Learning
1. INTRODUCTION
Social media has become a powerful tool for sharing
opinions. There are social media platforms like Facebook,
Twitter and Google+ to share opinions, reviews and
ratings. All major political parties and their members all
over the world have their official accounts on Twitter with
millions of followers. They consider this platform as a
medium to connect with young people who might vote
them. With significant rise of Indian users on Twitter
during the pandemic, people have been more vocal to
criticize or appreciate a political decision.
Sentimental Analysis is a method to teach a machine to
extract emotion from a given text [1]. A text can be
anything, a simple review, a social statement, tweets or
messages. Twitter Sentiment Analysis of tweets regarding
elections can be used by the general public as well as the
political parties to understand the positive or negative
views of people regarding a particular political party, thus,
helping to predict the election results during that period.
Elections play an important role in a democratic country.
Indian parliamentary system gives its people the right to
decide who will govern them for the next five years.
During the tenure of Feb 22 to March 22, five state
elections are lined up, with the important one being at
Uttar Pradesh, which sends the largest number of MPs to
parliament. The major national political parties contesting
in the elections are Bhartiya Janata Party(BJP), Indian
National Congress (INC), Aam Aadmi Party(AAP),
Samajwadi Party(SP), Shiromani Akali Dal(SAD) and Naga
People’s Front(NPF).
2. LITERATURE SURVEY
Parul S and Teng-Sheng Moh [2] predicted the results of
2016 Indian general elections using tweets in Hindi
language. They performed text mining on 42,235 tweets
collected over a month. They applied three ML algorithms.
The accuracy of the Naïve Bayes’ algorithm was 62.1% and
the accuracy of Support. Vector Machine was 78.4%. Final
prediction was done by utilizing SVM, since the accuracy
was higher.
Dr D. Rajeswara Rao and team [3] gathered a dataset of
more than 500,000 tweets out of which 80% was used for
training and rest for testing. They predicted which
political party had more influence on social media.
Proposed a system that trained the dataset for more than
2 days and a classifier was built. Experiments proved that
SVM was the most accurate model built with an accuracy
of 80%.
Ferdin Joe and John Joseph [4] used decision tree to
predict 2019 Indian General Elections. The results
obtained in the proposed methodology showed it had a
promising future in predicting Indian election results.
Meng-Hsiu Tsai and his team [5] at Middle Georgia State
University presented a machine learning strategy to
analyse Twitter data for predicting the results of local
elections in US. They categorized their results into 5
classes namely very positive, positive, neutral, negative
and very negative. They used RNTN model to calculate
weighted sentiment scores.
Payal Khurana Batra and her team [6] predicted election
results of Lok-Sabha 2019. After pre-processing, they split
the data into two parts containing BJP and Congress
tweets in separate sets. They trained their model using

five different ML algorithms. Decision tree and XGBoost
gave higher accuracy above 80%.
3. PROPOSED METHODLOGY
The proposed methodology can be implemented in five
phases. The first two phases are done periodically,
followed by prediction phase.
3.1 Data Collection
Tweets used for training the dataset were collected during
the period of November-December, 2021. A total of nearly
12000 tweets were collected. Different hashtags and
phrases like ’UPelection’, ’Punjabelection’, ’Yogi
Adityanath’, ’BJP elections 2022’, ’INC Punjab elections’,
etc. were used. Thus, a domain-specific corpora was
created and other features like ’like count’, ’retweet count’,
’user name’ and date-time of tweet’ were also included in
the dataset. Datasets of top political parties for every
Indian state(contesting in elections-2022) were collected
every 5 days during Jan-Feb period for testing the model.
Top three political parties considered for each state,
according to OneIndia [7] opinion polls, is shown in table-
1.
Table -1: Top political parties of each state
State Name Top 3 political parties
considered
Uttar Pradesh BJP, SP, INC
Punjab AAP, INC, SAD
Uttarakhand BJP, INC, AAP
Goa BJP, INC, AAP
Manipur BJP, INC, NPF
Two important libraries used were:
1. Tweepy : It is provided by Twitter. A collection of latest
as well as popular tweets of a particular hashtag were
collected and combined together.
2. Snscrape : As tweepy has a restriction on the amount of
tweets to be extracted and tweets older than 7 days
cannot be extracted, snscrape was used to overcome these
limitations.
3.2 Data Preprocessing
Text preprocessing is an important phase in ML workflow.
It essentially involves cleaning of the data to extract only
the meaningful information. First step towards data
cleaning is to ensure there is no duplicate or irrelevant
data. We acquired data using different hashtags over
different periods of time. It can often lead to redundant
tweets by the same or different users having multiple
common hashtags. We eliminated duplicity in data
collection phase itself. Next steps for preprocessing done
on text data are as follows:
1. Use of regular expressions:
We used regular expressions to remove website URLs,
replace ‘@handles’ with ‘handles’, ‘#hashtags’ with
‘hashtags’, and multiple spaces with single space. We also
removed special characters and punctuations.
2. Removal of stopwords:
A stopword is a commonly used word such as ‘the’, ‘a’, ‘an’,
‘in’, etc. These words do not add any meaning to the
sentence and merely used as a filler. The frequency of
these words are very high. We would not like these words
to take more space in our database and increase data
processing time.
3. Lemmatization:
It is important that all words in the corpora are in their
root or dictionary form, known as lemma. We do not want
our model to consider two words, having same contextual
meaning, as two different words. For example, the words
‘winning’, ‘winner’, ‘wins’ are all converted to their root
form ‘win’.
3.3 Labelling the dataset
After preprocessing, it is important to label the dataset.
We used VADER(Valence Aware Dictionary and sEntiment
Reasoner) [8] library to label the dataset into positive,
negative and neutral tweets. It uses lexical and rule-based
analysis to label the dataset. We use the compound value
of the polarity score to get the sentiment behind the tweet.
Table-2 shows the mapping of score to sentiment.
Table -2: Labelling the VADER compound score
Compound score Sentiment
>=0.05 Positive
>=-0.05 and <=0.05 Neutral
<=-0.05 Negative
3.4 Model Training
For the proposed work, the data was split into
training(0.75) and test data(0.25) and the feature
extraction technique used was tf-idf. The tfidf technique[9]
multiplies term frequency and inverse document

frequency, which provides a numerical value, denoting the
weight of a particular word in the document. The rare
words have higher tfidf value and considered important to
model training.
Further, supervised machine learning algorithms are used
to build a classification model. We used Logistic
Regression, Support Vector Machine and Random Forest
Classifier to predict the final output. We also used
combination of the above algorithms using ensemble
voting techniques. . We created a pipeline of tfidf along
with ML algorithms while training our model.
 Logistic Regression:
It is mainly used when output is a categorical variable.
It is an extension of Linear Regression, but instead of
fitting the regression line, we fit an S-shaped function
which saturates all the values between 0 and 1. It
gives the probability of the predicted output class.
 Support Vector Machine:
It is applicable when we want to classify an n-
dimensional space of datapoints having multiple
classes using a decision boundary called hyperplane.
The dimension depends upon the number of features.
 Random Forest Classifier:
It uses a combination of n decision trees to build a
classification model. It averages the accuracies of all
predictions of n trees. Since, it is an ensemble
technique, it has the ability to outweigh other
supervised algorithms in most of the cases.
 Voting Classifier:
We used a combination of the above algorithms to
build an voting classification model. We used both,
hard voting(selects a model using majority voting
technique) as well as soft voting(selects a model by
calculating the probability of each class and averaging
it), to build two different ensemble models.
By comparing the accuracies of the above models shown in
table-3, we found out Random Forest Classifier provided
better results than others. Thus, we used it to make
further predictions on unlabeled dataset.
Table -3: Model Analysis
Model Accuracy Score
Random Forest Classifier 77.59%
Voting Classifier(Soft) 74.69%
Logistic Regression 74.22%
Voting Classifier(Hard) 73.86%
Support Vector Machine 73.28%
3.5 Model Predictions
We created separate datasets for each political party for
each state. We applied the model on each of these datasets
to get the sentiment behind the tweet. In order to get the
popularity of a political party and extrapolate their
chances to win the State elections, we calculated the
popularity score, given by,
Popularity_score = (sum(tweets with positive sentiment) -
sum(tweets with negative sentiments) / (Total tweets over
the period))*100
The above score can also be called as ‘Effective positive
Rate’. We ranked and visualized the political parties by the
above score. We also calculated the percentage of positive,
negative and neutral tweets for all the parties. To provide
further analysis, we provided a timeline of tweets,
providing sentiments, for every party during the
campaigning period(Jan-Feb 2022) of election.
4. RESULTS AND DISCUSSIONS
The ‘Effective Positive Rate’ and comparison of percentage
of positive, negative and neutral tweets of top three
political tweets of Uttar Pradesh and Punjab are shown
chart-1 and chart-2 respectively.
Chart -1: Uttar Pradesh Analysis

BJP having 34.84% positive tweets and popularity score of
22.93% has higher chances of winning the UP state
elections. AAP, on the other hand, is predicted to be the
winner of Punjab elections, with effective positive rate of
22.37%.
Chart-3 provides in-depth analysis of BJP in Uttar Pradesh,
showing timeline of tweets of different sentiments. The
above figures show most of the tweets have neutral
sentiment, since most of the tweets are news articles and
political events of a particular party. Similarly, ranking of
popularity, over Twitter, is done for all the five states.
Chart -2: Punjab Analysis
Chart -3: Uttar Pradesh-BJP: Timeline of tweets
From table-4, it is clear that the political party, BJP has
dominated in majority of states. The actual results are
obtained from OneIndia.com[10] results. The observations
shows correct predictions for all states, except Manipur.
Since, the penetration of social media and internet is low
in Manipur among other states, thus leading to few
discussions and low buzz of Manipur elections on Twitter,
the results of this state cannot be rightly predicted by
Twitter. Therefore, the proposed paper shows that Twitter
as a platform can be effectively used as an election result
indicator for majority of Indian states.
Table -4: Actual and Predicted Winner
State Name Predicted winner Actual winner
Uttar Pradesh BJP BJP
Punjab AAP AAP
Uttarakhand BJP BJP
Goa BJP BJP
Manipur INC BJP
5. FUTURE WORK AND LIMITATIONS
The proposed system does not consider geographical
location of the tweet as a filter for state elections, as
Twitter does not provide adequate information about
user’s location, thus, a general mood of entire blogosphere
is considered to predict the election results. This work can
further be extended on tweets of different regional
languages of Indian states other than English to improve
accuracy. Currently, regional languages supported by
Twitter are Hindi, Gujrati, Marathi, Urdu, Tamil, Bengali,
and Kannada.
Sarcasm was not detected in some sentences due to misuse
of the semantics and not everyone has access to social
media where they can stand out from the crowd and
express their support for one another, are some of the
limitations of the proposed model.
6. CONCLUSION
An effective Random Forest classification model, with an
accuracy of 77.59%, was built to predict the popularity of
political party. The proposed system can be used by
political parties to improve their campaigning strategies
during the election period. It can be used by them as a part
of social media analytics to study the trends of other
political parties as well. User can make informed decision
in voting by seeing the current trends of political parties.
Political analyst and strategist can use this methodology, as
application, as a long term plan for a political party to study
the sentiments of people over a long time period.
Observing the expanded use of social media platforms, this
project concentrated on exploring of social platform
(Twitter) as the chase for elections’ campaign.

REFERENCES
[1] DataRobot, "Introduction to Sentiment Analysis: What
is Sentiment Analysis?," DataRobot, 26 March 2018.
[Online]. Available:
https://www.datarobot.com/blog/introduction-to-
sentiment-analysis-what-is-sentiment-analysis/.
[2] T.-S. M. Parul Sharma, "Prediction of Indian Election
Using Sentiment Analysis," 2016 IEEE International
Conference on Big Data (Big Data), pp. 1966-1971,
2016.
[3] S. U. S. K. M. S. R. G. C. U. J. Dr D Rajeswara Rao, "Result
prediction for political parties using twitter sentiment
analysis," International Journal of Computer
Engineering and Technology, no. 11(4), 2020.
[4] F. J. J. Joseph, "Twitter Based Outcome Predictions of
2019 Indian General Elections Using Decision Tree,"
2019 4th International Conference on Information
Technology (InCIT), Bangkok, THAILAND, pp. 50-53,
2019.
[5] Y. W. M. K. a. N. R. Meng-Hsiu Tsai, "A machine
learning based strategy for election result prediction,"
2019 International Conference on Computational
Science and Computational Intelligence (CSCI), pp.
1408-1410, 2019.
[6] A. S. S. a. C. G. Payal Khurana Batra, "Election Result
Prediction Using Twitter Sentiments Analysis," 2020
Sixth International Conference on Parallel, Distributed
and Grid Computing (PDGC), pp. 182-185, 2020.
[7] "Uttar Pradesh Assembly Election 2022 Opinion Poll,"
Oneindia, 2022. [Online]. Available:
https://www.oneindia.com/uttar-pradesh-election-
2022-opinion-poll-and-exit-poll/.
[8] GeeksforGeeks, 7 Oct 2021. [Online]. Available:
https://www.geeksforgeeks.org/python-sentiment-
analysis-using-vader/.
[9] Mamun, "Medium," 20 June 2019. [Online]. Available:
https://medium.com/@imamun/creating-a-tf-idf-in-
python-e43f05e4d424. [Accessed 20 May 2022].
[10] "Indian Elections 2022," Oneindia, 25 March 2022.
[Online]. Available:
https://www.oneindia.com/elections/.

Election Result Prediction using Twitter Analysis

More Related Content

Similar to Election Result Prediction using Twitter Analysis (20)

More from IRJET Journal (20)

Recently uploaded (20)

Election Result Prediction using Twitter Analysis