Handling Class Imbalance for Indonesian Twitter Sentiment Analysis A Comparative Study of Algorithms
Journal of System and Management Sciences · 2024
This research investigates the Multinomial Naive Bayes (MNB) and Logistic Regression (LR) algorithms for sentiment analysis on Indonesian language tweets related to ChatGPT.Before classification, TF-IDF and SMOTE will be implemented.A total of 16500 tweets written in the Indonesian language were collected.These tweets were subsequently classified using an Indonesian dictionary of positive and negative phrases.Subsequently, the process of case folding, data purification, tokenization, removal of stopwords, and stemming is executed.The SMOTE oversampling approach is employed to address the issue of class imbalance in the dataset.Comparative evaluation on a trial split of 80:20 shows that LR has a higher accuracy of 86% compared to MNB (74%).LR also shows superior precision, recall, and F1 scores.The results show better LR for Twitter sentiment analysis without significant improvement of the sampling technique.