Twitter Sentiment Analysis on the Iran-Israel Conflict Using the Naïve Bayes Classification Algorithm

Abstract

The armed conflict between Iran and Israel, which has attracted global attention, has sparked various public reactions, including from the Indonesian community. Given its potential impact on global social and economic stability, it is important to systematically analyze public perceptions using a sentiment analysis approach. A total of 310 tweets were collected through a crawling process and processed using several preprocessing stages, such as text cleaning, normalization, stopword removal, tokenization, stemming, and translation. Labeling was performed directly using the Naive Bayes algorithm, by comparing three algorithms: Gaussian Naive Bayes, Multinomial Naive Bayes, and Bernoulli Naive Bayes. Performance evaluation was conducted using metrics such as accuracy, precision, recall, and F1-score. The classification results showed that Multinomial Naive Bayes achieved an accuracy of 75.81%, Gaussian Naive Bayes achieved 77.42%, while Bernoulli Naive Bayes achieved 87.1%. Bernoulli Naive Bayes demonstrated superior performance in handling textual data with word frequency representation. This study contributes to strengthening the use of machine learning methods for public opinion analysis on social media, particularly in the context of geopolitical issues. The findings indicate that Bernoulli Naive Bayes is more suitable for classifying public opinion texts compared to the Gaussian and Multinomial variants.