Journal of Artificial Intelligence and Engineering Applications

Website: https://ioinformatic.org/

15th February 2024. Vol. 3. No. 2; e-ISSN : 2808-4519

Classification of Public Sentiment Toward 2024 Presidential

Candidates on Social Media Platform X Using Naïve Bayes
Algorithm
Ramdhan Hakiki1*, Agung Pambudi2, Asriyanik3
1,2,3
Teknik Informatika, Fakultas Sains dan Teknologi, Universitas Muhammadiyah Sukabumi
[email protected] 1, [email protected] 2, [email protected] 3

Abstract

This research examines the use of Naïve Bayes algorithm to classify public sentiment on social media X towards Indonesia's 2024
presidential candidates. Against the backdrop of the importance of presidential elections in a democracy, this research focuses on analyzing
public sentiment from June to August 2023. The Naïve Bayes method was chosen to process review data about the three main candidates.
The classification results provide insight into the positive and negative sentiments of the public, providing benefits for political parties and
researchers in understanding public opinion. This research also enhances the understanding of sentiment classification in a political context
and provides readers with a useful reference on the Naïve Bayes approach to sentiment classification. In terms of accuracy, the developed
naïve bayes model shows a success rate with an accuracy of 74% for Anies Baswedan, 74% for Ganjar, and 88% for Prabowo.

Keywords: Sentiment Classification, Naïve Bayes, Presidential Candidates

1. Introduction
General elections have been held many times in Indonesia's history, but direct elections by the Indonesian people began for the first time
during the reform period after the collapse of the New Order era, namely in 2004. The 2024 Indonesian presidential and vicepresidential
elections, also known as the 2024 presidential election, is a democratic stage to elect the president and vice president of the republic of
Indonesia for the 2024-2029 service period, which will be held on Wednesday, February 14, 2024. The presidential election is a very
important political event in a country. The election will be the fifth direct presidential and vicepresidential election in Indonesia. According
to the 1945 Constitution, the president and vice president are elected in one pair and filled by direct election by the people in article 6A
paragraph (1) of the third amendment of the 1945 Constitution "The president and vice president are elected in one pair and filled by direct
election by the people" [1].
Looking at public opinion about the 2024 presidential candidate, it is not lost on the attention expressed by the public on social media, one
of which is social media platform X related to the presidential candidate in the 2024 presidential election. Twitter, which is now social
media platform X, has seen its users in Indonesia grow rapidly in recent years. Social media X is very popular in Indonesia with millions
of active users who use social media X to interact, express opinions and also share or search for information with current news and trends
[2].
According to [3]sentiment classification is a method to find out the opinion of a person or group of people on certain issues, products,
services or groups. Sentiment classification is the process of identifying and categorizing views or feelings represented in text or written
language into positive, negative, or neutral sentiment categories.
There are several methods that can be used in performing sentiment classification, including K-Nearest Neighbor (KNN), Support Vector
Machine (SVM) and also Naïve Bayes. Previous research on sentiment analysis of online loans on twitter used the Support Vector Machine
(SVM) method which resulted in an accuracy of 62% [4]. Another study, namely sentiment analysis of perceptions of the elected
government in the 2019 presidential election on twitter using Naïve Bayes, resulted in an accuracy value of 81% [5]. And other research
using the K-Nearest Neighbor method with the title "sentiment analysis of twitter users on the Indonesian football polemic using TF-IDF
weighting and K-Nearest Neighbor" with an accuracy value of 79.99% [6].
With the comparison of methods for sentiment classification above, researchers will use the Naïve Bayes algorithm because it obtained the
largest accuracy value with 81%. According to [7], Naïve Bayes is a classification method that is widely used in processing data mining
or text mining. By applying the Naïve Bayes technique to sentiment classification with case studies of opinions or reviews of the 2024
presidential candidates on social media.
552 Journal of Artificial Intelligence and Engineering Applications

2. Research methods
This research method aims to describe the stages of the method and research design based on data collection or information obtained, as
well as the research tools to be used and the schedule of research activities to be carried out. The implementation of this research also aims
to analyze public sentiment towards presidential candidates on the X- social media platform using the Naïve Bayes algorithm. The stages
of this research method are as follows:

Data selection Preprocessing dataset Transformation

n

Visualization Evaluation Modeling

Fig. 1: Stages of Sentiment Classification in research

In this research method, the Knowledge Discovery in Database (KDD) method is used by researchers to apply the Naïve Bayes algorithm
to process and analyze tweet data. This method is used with the aim that researchers can extract valuable information hidden in the review
data set so that previously unknown, useful and potential knowledge can be found for future decision making. Based on the previous
Knowledge Discover In Database (KDD) explanation, here are the steps for the KDD method:

2.1. Data selection

Data Selection is the initial stage of this research which will select and take reviews from social media user X-'s tweets regarding the 2024
presidential candidates and the process is collected using data crawling techniques via the Twitter API using Google Colaboratory. Data
collected on Presidential Candidate Prabowo Subianto amounted to 1000 tweets, Ganjar Pranowo 1000 tweets and Anies Baswedan 1000
tweets from June 2023 to August 2023 which will then be processed again according to the sentiment classification stage.

2.2. Pre-processing

The pre-processing dataset stage is the stage of processing the dataset which was originally in the form of raw data or dirty data which will
be processed into data according to needs for further classification because in this preprocessing there are many stages that will be carried
out to process the data, namely cleaning, casefolding, tokenizing, filtering, slang words replacement and stemming. The aim of this stage
is also to produce quality and relevant data to carry out the desired sentiment classification process.

2.3. Transformation

The transformation stage is part of the KDD process which aims to change or modify review data so that it is easier to process and analyze.
This stage is usually carried out after the dataset preprocessing stage and can include various processes, such as word weighting using
Term Frequency inverse document frequency (TF-IDF). By carrying out transformations, tweet data will be easier to process and analyze
so that the analysis results are more accurate and useful.

2.4. Modeling

The modeling stage is a stage in the Knowledge Discovery in Database (KDD) process which aims to build a model that is able to apply
patterns or relationships to review data. In this step, the training data that has been previously processed and transformed will be used to
build the model. Many algorithms can be used in the modeling stage, one of which is Naïve Bayes.

2.5. Evaluazation

The evaluation stage in Knowledge Discovery in Database (KDD) is the stage of assessing the results of the model that has been built at
the modeling stage. In this step the model has been built and will be applied to the test data that has been prepared previously. The results
of applying the model will be analyzed and compared with the expected results, so that the level of accuracy of the model can be known.
There are several metrics that can be used in evaluation, such as accuracy, precision, recall, and f1-score. Apart from that, the confusion
matrix can also be used to find out how well the model can predict actual results. It is important to carry out this evaluation stage in order
to find out to what extent the model being built can capture the patterns or relationships contained in the tweet data.

2.6. Visualization

The visualization stage in Knowledge Discovery in Database (KDD) is the final stage of the KDD process which aims to display the results
of the model that has been built in the previous stage in a visual form that is easy for other people to understand. After the tweet data has
been successfully processed in the previous step, the next step is to display it in the form of a diagram or graph, making it easier to interpret
and analyze the tweet data. Some examples of visualizations that can be used in KDD are scatter plots, bar charts or wordclouds. It is
important to carry out this visualization stage so that it can make it easier to make decisions based on the results of the model that has been
created.
Journal of Artificial Intelligence and Engineering Applications 553

3. Result and discussion

3.1. Data selection

Review data was obtained from social media platform X using search keywords such as #capres2024, #prabowo, #anies, and #ganjar. The
data is in the form of tweets in Indonesian that were crawled from social media X, totaling 1000 tweets from each presidential candidate.
After that, the data will be saved in CSV format.

3.2. Preprocessing dataset

Next is the pre-processing stage, which is the process of preparing data before it is applied to machine learning models or algorithms. The
goal is to clean, format, and prepare the raw data so that it can be analyzed in modeling. The following are the stages of preprocessing in
this research.

3.2.1. Cleaning

The first step is to clean the dataset to remove symbols, emoticons, or noise words in the tweet data.

Table 1: Cleaning
Appearance
Input Output
Semangat pak…semoga bisa memimpin negeri ini Semangat pak semoga bisa memimpin negeri ini

3.2.2. Case folding

Case folding is a step to replace or change the entire case of the text to lowercase because in the dataset of tweets of 2024 presidential
candidates there are uppercase and lowercase letters. Therefore, a case folding step is needed so that all words used are lowercase.

Table 2: Case folding

Appearance
Input Output
Semangat pak semoga bisa memimpin negeri ini semangat pak semoga bisa memimpin negeri ini

3.2.3. Tokenizing

At this stage, tokenization is used to break sentences into words, on a space-separator basis.

Table 3: Tokenizing
Appearance
Input Output
‘semangat’ ‘semoga’ ‘bisa’ ‘memimpin’ ‘negeri’ ‘semangat’ ‘pak’ ‘semoga’ ‘bisa’ ‘memimpin’ ‘negeri’ ‘ini’

3.2.4. Slang words replacement

The process of consistent terms that have the same meaning but are written differently may be necessary, due to misspellings, use of
abbreviations, or informal language.

Table 4: Slang words replacement

Appearance
Input Output
krg kurang

3.2.5. Stopword

The next step is to implement stopwords, which aims to remove words that have no meaning (stoplist) or have no effect on accuracy in the
classification process, such as linking words. The results of the stopword process can be checked in the following table.

Table 5: Stopword
Appearance
Input Output
‘semangat’ ‘pak’ ‘semoga’ ‘bisa’ ‘memimpin’ ‘negeri’ ‘ini’ ‘semangat’ ‘semoga’ ‘bisa’ ‘memimpin’ ‘negeri’

3.2.6. Stemming

Stemming in this pre-processing step is done to convert words into their base word form by removing the affixes attached to the word.

Table 6: Stemming
Appearance
Input Output
554 Journal of Artificial Intelligence and Engineering Applications

‘semangat’ ‘semoga’ ‘memimpin’ ‘negeri’ ‘semangat’ ‘semoga’ ‘pimpin’ ‘negeri’

3.3. Tf-Idf

In this research, the TF-IDF technique, which is a term weighting method, is applied to assign a value or weight to each word in visitor
reviews. This facilitates the classification process. Initially, the word frequency in a document is calculated, followed by the calculation of
the inverse document frequency. The results of the Term Frequency and Inverse Document Frequency calculations can be seen in Table 7.

Table 7: The result of calculating TF and calculating IDF

Appearance
Term Tf Idf
semangat 1 0.90308
semoga 1 0.90308
pimpin 1 0.90308
negeri 1 0.90308

3.4. Implementation of naive bayes

The implementation of classification with the Naïve Bayes method is carried out using data that has been processed through the text
preprocessing stage and has been weighted using the TF-IDF method. The implementation of performing the classification stage using
naïve bayes consists of 2 stages as follows:

3.4.1. Naïve Bayes modeling

The first step in performing classification is modeling using the Naïve Bayes method. This process begins with the creation of a Naïve
Bayes model using the sklearn library in the Python programming language, specifically using the multinomial Naïve Bayes method.
Afterwards, the classification model is formulated using the fit function to train the model using the training data. The following is the
modeling program code:

from sklearn.naive_bayes import MultinomialNB

#pembentukan model Naïve Bayes
clf=MultinomialNB()
model = clf.fit(vec_latih,y_train)

Fig. 2: Program code for naïve bayes modeling

3.4.2. Test data prediction

The data set previously separated by 20%, referred to as x_test in the implementation code, will be used to test the data. This testing process
is carried out using the Naïve Bayes model that has been created previously. The following is the program code to test the data:

#term frequency(pembobotan kata) data uji

vec_uji=vec.transform(x_test)
vec_uji
#klasifikasi data uji
predicted=clf.predict(vec_uji)

data_uji.insert(2, column='label_bayes',
value=predicted)
data_uji

Fig. 3: Program code for data testing

After passing the data testing stage using the test data as described earlier, the results can be seen in the "hasil_stemming" column including
the test data, while "label" refers to the initial label of the test data. Meanwhile, the "label_bayes" column displays the classification results
of the data that has been tested using the prepared model. The result display is as follows:
Journal of Artificial Intelligence and Engineering Applications 555

Fig. 4: Data testing results using naïve bayes classifier

3.5. Evaluation

The performance evaluation of the program model in this study involves the use of confusion matrix. In this evaluation stage, the goal is
to assess the performance of the algorithm used by calculating the accuracy, recall, precision, and f1-score values. The implementation of
evaluation on test data to measure the performance of the Naïve Bayes algorithm model that has been developed is implemented using the
sklearn metrics and seaborn libraries. The data results for the sentiment classification process using confusion matrix can be found in Table
8.

Table 8: Confusion matrix prabowo subianto

Positif Negatif
Positif 48 4
Negatif 21 126

Table 9: Confusion matrix anies baswedan

Positif Negatif
Positif 33 6
Negatif 54 106

Table 10: Confusion matrix ganjar pranowo

Positif Negatif
Positif 10 0
Negatif 44 145

Next to display the accuracy, precision, recall and f1score values by running the program code as follows:

from sklearn.metrics import

accuracy_score,precision_score,recall_score,f1_score
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix

target_names = ['Positif', 'Negatif']

print('=========================================\n')
print(classification_report(y_test,predicted,
target_names=target_names))

Fig. 5: Program code for finding model performance values

The results of the above program code for each dataset can be seen as follows:
Table 11: Performance value for each dataset
Precision Recall F1-score
Dataset
Positif Negatif Postif Negatif Positif Negatif
Anies baswedan 0.85 0.66 0.38 0.95 0.52 0.78
Ganjar pranowo 1.00 0.77 0.19 1.00 0.31 0.87
Prabowo subianto 0.92 0.86 0.70 0.97 0.79 0.91

4. Conclusion
This research successfully uses the Naïve Bayes algorithm to classify public sentiment towards 2024 presidential candidates on X social
media. The results demonstrate the effectiveness of this algorithm in the context of sentiment classification, providing insights to political
parties and presidential candidates. Through Knowledge Discovery in Databases (KDD) methodology, the research identified public
perceptions of Prabowo (64% positive, 36% negative), Anies Baswedan (50.9% positive, 40.1% negative), and Ganjar (74.2% positive,
25.8% negative). The Naïve Bayes model achieved significant accuracy, namely 74% for Anies Baswedan, 74% for Ganjar, and 88% for
Prabowo. In conclusion, this algorithm is useful in understanding public views on political figures, providing a basis for political parties
and presidential candidates to design strategies based on social media responses.
556 Journal of Artificial Intelligence and Engineering Applications

References
[1] J. T. Nugraha and UUD, “No Analisis struktur kovarians indikator terkait kesehatan pada lansia yang tinggal di rumah, dengan fokus pada rasa
subjektif terhadap kesehatan Title,” vol. 105, no. 3, pp. 129–133, 1945, [Online]. Available:
https://webcache.googleusercontent.com/search?q=cache:BDsuQOHoCi4J:https://media.neliti.com/media/publications/9138-ID-perlindungan-
hukum-terhadap-anak-dari-konten-berbahaya-dalam-media-cetak-dan-ele.pdf+&cd=3&hl=id&ct=clnk&gl=id
[2] Rangga, “No Title,” kompas.id, 2023. https://www.kompas.id/baca/riset/2023/12/14/media-sosial-pengaruhi-pemilih-pada-pemilu-2024
[3] S. Suryono, E. Utami, and E. T. Luthfi, “Klasifikasi Sentimen Pada Twitter Dengan Naive Bayes Classifier,” Angkasa J. Ilm. Bid. Teknol., vol.
10, no. 1, p. 89, 2018, doi: 10.28989/angkasa.v10i1.218.
[4] D. S. Utami and A. Erfina, “Analisis Sentimen Pinjaman Online di Twitter Menggunakan Algoritma Support Vector Machine (SVM),” SISMATIK
(Seminar Nas. Sist. Inf. dan Manaj. Inform., vol. 1, no. 1, pp. 299–305, 2021.
[5] F. A. Wenando, R. Hayami, and A. J. Anggrawan, “Analisis Sentimen Pada Pemerintahan Terpilih Pada Pilpres 2019 Ditwitter Menggunakan
Algoritme Naïvebayes,” JURTEKSI (Jurnal Teknol. dan Sist. Informasi), vol. 7, no. 1, pp. 101–106, 2020, doi: 10.33330/jurteksi.v7i1.851.
[6] J. A. Septian, T. M. Fachrudin, and A. Nugroho, “Analisis Sentimen Pengguna Twitter Terhadap Polemik Persepakbolaan Indonesia Menggunakan
Pembobotan TF-IDF dan K-Nearest Neighbor,” J. Intell. Syst. Comput., vol. 1, no. 1, pp. 43–49, 2019, doi: 10.52985/insyst.v1i1.36.
[7] R. L. Atimi and Enda Esyudha Pratama, “Implementasi Model Klasifikasi Sentimen Pada Review Produk Lazada Indonesia,” J. Sains dan Inform.,
vol. 8, no. 1, pp. 88–96, 2022, doi: 10.34128/jsi.v8i1.419.