JUSIKOM PRIMA (Jurnal Sistem Informasi dan Ilmu Komputer Prim.
Vol.
6 No.
Agustus 2022
E-ISSN : 2580-2879
NAyaVE BAYES ALGORITHM OPTIMIZATION USING PARTICLE
SWARM OPTIMIZATION (PSO) FOR COVID-19 VACCINE
SENTIMENT ANALYSIS ON TWITTER
Rivan Adi Nugraha1.
Teguh Iman Hermanto2.
Imam MaAoruf Nugroho3
1,2,3
Sekolah Tinggi Teknologi Wastukancana Jl.
Cikopak No.
Sadang.
Purwakarta - Jawa Barat - Indonesia E-mail : rivanadi82@wastukancana.
ABSTRACT- The Covid-19 vaccine is a vaccine that is quite popular, because it is the most needed and most discussed vaccine.
There are 5 types of vaccines that are very popular including AstraZeneca.
Moderna.
Pfizer.
Sinopharm and Sinovac.
Sentiment analysis is a branch of text classification with computational linguistics and natural language processing that refers to a broad field, and text mining has a function to analyze opinions, judgments, sentiments, attitudes, evaluations and emotions of a person regarding an individual, organization, certain topics, services and other activities.
This study aims to classify public sentiment towards the type of Covid-19 vaccine on social media Twitter, whether the opinion is positive or negative by using the Nayve Bayes algorithm based on Particle Swarm Optimization (PSO).
The conclusion of this study is that the results of testing the Nayve Bayes algorithm with PSO using RapidMiner software are 79.
17% accuracy, 87.
69% precision, 07% recall for AstraZeneca vaccine, 68.
82% accuracy, 92.
29% precision, 71.
72% recall for Moderna vaccine, 54% accuracy, precision 77.
83%, recall 62.
95% for Pfizer vaccine, accuracy 93.
33%, precision 91.
00% for Sinopharm vaccine, and accuracy 74.
93%, precision 82.
61%, recall 70.
90% for Sinovac It can be concluded that with the help of optimization PSO, the resulting confusion matrix value is greater and is proven to be more accurate.
Keywords : Vaccine.
Covid-19.
Sentiment Analysis.
Naive Bayes.
Particle Swarm Optimization.
PRELIMINARY
On March 11, 2020, the international health organization World Health Organization or WHO, officially determined the corona virus (Covid-.
to be a pandemic level which was initially still at the outbreak level .
At the end of 2019, even though the center of the spread of this virus was originally in China, precisely Wuhan City .
Currently, the virus is still infecting and spreading to all regions in Indonesia .
One solution to anticipate the spread of this virus is by developing a vaccine .
Vaccines are useful for protecting people who are vaccinated, and can reduce the rate of disease spread .
To stop and prevent the spread of disease, it is very important for us to develop effective and safe vaccines .
At this time the Covid-19 vaccine is a vaccine that is quite popular, because it is the most needed and most discussed vaccine.
Coming from abroad and domestically, there are quite a number of types of Covid-19 vaccines that make people confused because they have many choices.
So that people will seek information about the Covid-19 vaccine, possibly by reading opinions or reviews.
Opinions or reviews from the public can certainly provide considerable benefits for other communities, because the public can get information from other community reviews, the majority of which share their opinions and personal experiences about the Covid-19 vaccine through social media networks, especially on the Twitter platform.
Twitter is one of the social media platforms that can provide information in the form of images or reviews to the public .
There are so many opinions or reviews on Twitter, it will take a lot of time to read all the reviews one by one to find that Sentiment analysis is one way to solve the problem by grouping opinions or reviews into positive sentiments or negative sentiments.
Sentiment analysis is a branch of text classification with computational linguistics and natural language processing which refers to a broad field .
Text Mining has a function to analyze evaluations and emotions of a person regarding an individual, organization, topic, service and certain other activities .
Sentiment analysis is useful for determining whether the opinion or review has a positive or negative tendency .
Classification methods that are often used in sentiment analysis for text categories are Support Vector Machine (SVM) and Nayve Bayes .
The Naive Bayes algorithm or method is very suitable to be implemented on data whose scale is very large and is able to overcome missing values .
mpty data value.
The Naive Bayes method also has drawbacks, where the magnitude of the classification accuracy cannot be calculated from the probability .
Therefore, it is necessary to optimize with the help of the Particle Swarm Optimization (PSO) algorithm by assigning a weight value to each attribute to produce increased accuracy and assist the public in determining the best type of JUSIKOM PRIMA (Jurnal Sistem Informasi dan Ilmu Komputer Prim.
Vol.
6 No.
Agustus 2022
E-ISSN : 2580-2879
Covid-19 vaccine based on the results of a review from Twitter.
In this study, samples of 5 types of Covid-19 vaccines will be taken including AstraZeneca.
Moderna.
Pfizer.
Sinovac and Sinopharm.
The purpose of this study is to find out the results of the analysis of 5 types of Covid-19 vaccines on Twitter social media using the Naive Bayes algorithm and Particle Swarm Optimization (PSO), compare the level of accuracy obtained between Nayve Bayes and Nayve Bayes PSO, and produce visualizations.
data from the research results.
RESEARCH METHODS
In this study, there are several stages carried out including data collection, data cleaning, labeling, preprocessing text, classification, and evaluation.
2 Cleaning Data At this stage the process of deleting each line of tweet data containing statements or spam in the dataset, so that the dataset contains purely opinions.
3 Labelling At this labeling stage, the data labeling process is carried out to divide and determine the sentiment of the tweet data whether it contains positive or negative sentiment.
4 Preprocessing Text At this text preprocessing stage, several techniques are carried out to carry out preprocessing stages including transformation, tokenization, and Picture 2.
Text Preprocessing Stage Picture 1.
Research Stages 1 Data Collection At this data collection stage, it contains 3 further stages, which include Literature Study.
Crawling Data and Tweet Data.
In the literature study stage, it is done by searching, analyzing and reading references from various theoretical sources related to sentiment analysis using the Naive Bayes algorithm and Particle Swarm Optimization (PSO).
After that in the next stage, researchers must get the Twitter API Key first before crawling data on Twitter using Orange software.
Then, after the data crawling process is complete, the data can be saved in a csv file format with the extension.
5 Klasifikasi At this classification stage, to produce a text classification that contains positive and negative, the word weighting process is carried out first.
The algorithm used in this stage is Nayve Bayes which is optimized with Particle Swarm Optimization (PSO).
6 Evaluasi At this evaluation stage, the performance of the nayve bayes PSO algorithm was tested for sentiment analysis of the covid-19 vaccine on the Twitter platform.
JUSIKOM PRIMA (Jurnal Sistem Informasi dan Ilmu Komputer Prim.
Vol.
6 No.
Agustus 2022
E-ISSN : 2580-2879
RESULTS AND DISCUSSION
1 Crawling Data In this stage the data crawling process is carried out with the help of the Twitter API Key, and data crawling is carried out on April 14, 2022 using the Orange software.
The keywords used were AuastrazenecaAy.
AumodernaAy.
AupfizerAy.
AusinopharmAy and AusinovacAy which resulted in 412 AstraZeneca datasets, 2000 Moderna datasets, 1937 Pfizer datasets, 333 Sinopharm datasets and 1042 Sinovac datasets, with a total of total data is 5724 datasets.
The following are the steps of the data crawling process carried out using the Orange software.
the final result of this dataset contains purely The total number of tweet data after this process is done is 1363 data.
Table 1.
Total data after cleaning Dataset Total Data AstraZeneca Moderna Pfizer Sinopharm Sinovac 3 Labelling After the Data Cleaning process is carried out, then the Labeling process is carried out by adding a computerized AuSentimentAy attribute column.
Picture 6.
Added the AuSentimentAy attribute Picture 3.
The process of crawling data with Orange Entering Twitter API Key Picture 4.
Twitter API key and secret Enter keyword query Below is an example of the keyword entered, namely "astrazeneca" for the data search process.
The data search is carried out based on content or tweets in Indonesian with a maximum data collection of 2000 tweets.
The results of the data labeling process resulted in 355 positive sentiment data and 1008 negative sentiment data with the following details:
Table 2.
Results of labelling on the dataset Keywords Total Data Positive Negative AstraZeneca 87 Moderna Pfizer Sinopharm Sinovac 4 Preprocessing Text Text preprocessing is a process that is carried out with the aim of avoiding inconsistent data, imperfect data, and disturbances contained in the data .
In this process, the transformation, tokenization, and filtering stages will be carried out.
Transformation In this process, the lowercase and remove url stages are carried out.
That is the process to change the letter data of all text into lowercase .
and delete the url contained in the tweet data line.
Table 3.
Before the process Transformation cOS3 AstraZeneca, loyal don't switch to the next brand eE https://t.
co/DCJzHhzv8c Table 4.
After the process Transformation s3 astrazeneca, loyal don't switch to the next Picture 5.
Enter the keyword AuastrazenecaAy 2 Cleaning Data Cleaning data is done to delete each row of tweet data that contains statements or spam, so that JUSIKOM PRIMA (Jurnal Sistem Informasi dan Ilmu Komputer Prim.
Vol.
6 No.
Agustus 2022
E-ISSN : 2580-2879
Tokenization In this process, a sentence is broken or cut per word or a sequence of string pieces, so that the end result becomes a word from sentence fragments.
Table 5.
Tokenization Process s3 astrazeneca, loyal don't switch to the next donAot Filtering There are 2 stages in this process: stopwords and regexp.
The process is to delete words that often appear, but do not have a meaningful meaning and remove connecting words that have no effect .
Table 6.
Before the process Filtering donAot Table 7.
After the process Filtering Picture 8.
Nayve Bayes PSO Model 5 Classification In this classification stage, the method or algorithm used is Nave Bayes which is optimized by Particle Swarm Optimization (PSO).
Calculation of nayve bayes is done by calculating the simple probability of each class from all training data.
After that, it is continued with the testing process, which is to determine the accuracy of the model built in the training process.
The following is a model of Nayve Bayes and Nayve Bayes PSO using RapidMiner 6 Evaluation In this stage, a process is carried out to measure the values of accuracy, precision, and recall by using a confusion matrix.
The following is the result of the confusion matrix for each type of vaccine.
Table 8.
Results of confusion matrix AstraZeneca
Confusion Matrix
NB PSO
Accuracy 71,39% 79,17% Precision 80,88% 87,69% Recall 82,09% 85,07% Table 9.
Results of confusion matrix Moderna
Confusion Matrix
NB PSO
Accuracy 67,62% 68,82% Precision 90,76% 92,29% Recall 71,72% 71,72% Table 10.
Results of confusion matrix Pfizer
Confusion Matrix
NB PSO
Accuracy 60,94% 67,54% Precision 69,77% 77,83% Recall 59,76% 62,95% Picture 7.
Nayve Bayes Model Table 11.
Results of confusion matrix Sinopharm
Confusion Matrix
NB PSO
Accuracy 80,00% 93,33% Precision 86,36% 91,67% Recall 86,36% 100,00% JUSIKOM PRIMA (Jurnal Sistem Informasi dan Ilmu Komputer Prim.
Vol.
6 No.
Agustus 2022
E-ISSN : 2580-2879
Table 12.
Results of confusion matrix Sinovac
Confusion Matrix
NB PSO
Accuracy 68,04% 74,93% Precision 72,87% 82,61% Recall 70,15% 70,90% 7 Visualization Picture 13.
WordCloud visualization of all Vaccine CLOSING Picture 9.
Column diagram of the amount of data for each type of Vaccine Picture 10.
Column chart of the amount of data, positive sentiment and negative sentiment for each type of Vaccine 1 Conclusion Based on research that has been conducted on sentiment analysis of the Covid-19 vaccine type on Twitter social media using the PSO-based Nayve Bayes algorithm, the data used include 412 AstraZeneca data, 2000 Moderna data, 1937 Pfizer data, 333 Sinopharm data and 1042 Sinovac data.
The results of testing the Nayve Bayes algorithm based on PSO using RapidMiner software include:
17%, precision 87.
69%, recall 85.
on AstraZeneca vaccine, accuracy 68.
82%, precision 29%, recall 71.
72% on Moderna vaccine, 54%, precision 77.
83%, recall 62.
on Pfizer vaccine, 93.
33% accuracy, precision 67%, recall 100.
00% on Sinopharm vaccine, and 93%, precision 82.
61%, recall 70.
on Sinovac vaccine.
This shows that with the help of optimization PSO is proven to be able to produce more accurate confusion matrix values.
2 Suggestions The suggestions given from the results of this study are expected to be developed for further research, and experiments are carried out using algorithms and optimization such as Support Vector Machine (SVM).
Decision Tree, and so on to get different outputs.
BIBLIOGRAPHY