Public Sentiment Analysis on Instagram and X toward the Cancellation of Fuel Purchases from Pertamina by Private Fuel Stations Using the BERT Method

Abstract

This study aims to analyze public sentiment toward the cancellation of fuel purchases from Pertamina
by private fuel stations based on comments collected from Instagram and X (Twitter). Sentiment analysis was
conducted using the Bidirectional Encoder Representations from Transformers (BERT) model, which is capable
of capturing contextual semantic information in text more effectively than traditional machine learning
approaches.The research stages include data collection through web scraping, text preprocessing, sentiment
labeling into positive, neutral, and negative categories, fine-tuning a pretrained Indonesian BERT model, and
model evaluation using accuracy, macro recall, and F1-score. The experimental results indicate that the BERT
based model is able to classify public sentiment effectively. Negative sentiment dominates the dataset and is
characterized by expressions of dissatisfaction and criticism toward the policy. Neutral sentiment mainly contains
descriptive and informational statements, while positive sentiment appears in a considerably smaller proportion.
The findings provide insights into public perceptions of the policy related to Pertamina and private fuel stations
and demonstrate the potential of BERT-based sentiment analysis as a decision-support tool for monitoring public
opinion on social media.