A Comparative Study of Machine Learning Algorithm for Sentiment Analysis Using Word2Vec and Synthetic Minority Oversampling Technique (SMOTE) on COVID-19 Vaccination Program
Rahmatika Pratama Santi, Fajril Akbar, Febby P. M. Piter · 2024
COVID-19 is still ongoing and has not disappeared until now in the community. The government has made various efforts to overcome this case, one of which is the vaccination program. Many opinions were generated from the public regarding the vaccination program. Opinions that arise need to be processed to determine responses and provide feedback regarding the program, so an effective and efficient process is needed in conducting sentiment analysis regarding the COVID-19 vaccination program. The resulting sentiments are processed using Word2Vec and Synthetic Minority Oversampling Technique (SMOTE) by comparing the Support Vector Machine (SVM), Naïve Baye, and Extreme Gradient Boosting (XGBoost) algorithms in order to produce an optimal classification model. The data are classified into positive, neutral and negative classes. Word2Vec is used to produce word vector representations that are ready to be trained with a classification algorithm using SMOTE to overcome class imbalance. The research results show that XGBoost produces the best value by classifying the f1 score for negative data by 90% and positive data by 90%. Based on the training process carried out, it was found that XGBoost can be used to analyze sentiment with quite accurate results compared to Naïve Bayes and SVM.