Analisis Sentimen Timnas Indonesia pada Data Tidak Seimbang Menggunakan Perbandingan Naïve Bayes dan IndoBERT

Abstract

Social media platform X is widely used by the public to express opinions on the performance of the Indonesian National Team, especially in the fourth round of the 2026 World Cup Qualifiers. In this phase, the Indonesian National Team suffered two consecutive defeats, namely 2–3 to Saudi Arabia and 0–1 to Iraq, which triggered an increase in emotional responses and public criticism on social media. This condition makes sentiment analysis important to understand public perception more objectively. This study aims to analyze the sentiment of social media users X and compare the performance of the Naïve Bayes and IndoBERT models in imbalanced data conditions. The research data amounted to 1,268 tweets that were processed through a pre-processing stage, then automatically labeled using a lexicon-based approach as an initial labeling into two classes, namely positive and negative. The dataset was divided into training data and test data with a ratio of 70:30. The data distribution shows the dominance of negative sentiment at 84.1% and positive at 15.9%. Classification was performed using TF-IDF-based Naïve Bayes and IndoBERT-base-p1, with data imbalance management using random oversampling and class weighting. The results show that Naïve Bayes without treatment achieved 84% accuracy but failed to recognize the positive class. After oversampling, the positive class recall increased to 45%. IndoBERT achieved 85% accuracy, with positive recall increasing from 35% to 43% and the positive class F1-score increasing by 47% after applying class weighting. Despite the relatively high accuracy, the evaluation shows the importance of considering performance on minority classes. Overall, IndoBERT with class weighting provided more balanced results. However, the use of lexicon-based automatic labeling is a limitation of this study.