ANALISIS SENTIMEN OPINI PUBLIK TERHADAP KASUS KORUPSI BAHAN BAKAR MINYAK OPLOSAN PT PERTAMINA DENGAN HYBRID MODEL DEEP LEARNING

Abstract

The corruption case related to oplosan fuel oil involving PT Pertamina has become a national issue that has drawn diverse responses from the public. Sentiment analysis of public opinion on social media can provide important insights for the government and stakeholders in understanding public perceptions of the case. This study aims to analyze public opinion sentiment regarding the alleged fuel adulteration corruption case involving PT Pertamina, using a hybrid deep learning model approach. Data were collected from the social media platform Twitter (X) between February 24 and March 19, 2025, resulting in 12,365 tweets after preprocessing. The study implements four model architectures: IndoBERT, CNN, LSTM, and a hybrid IndoBERT-CNN-LSTM model. Evaluation results show that IndoBERT achieved the highest accuracy at 90%, followed by CNN (86%), hybrid (84%), and LSTM with the lowest accuracy (69%). In addition, the K-Fold cross-validation scheme produced more stable model evaluation results than the Hold-Out method. Based on sentiment distribution analysis, public opinion was dominated by negative sentiment at 72%, while positive and neutral sentiments each accounted for 16%. These findings indicate that the public tends to respond negatively to the Pertamina fuel corruption issue. This study contributes to the understanding of public opinion on social media through a deep learning-based sentiment analysis approach and highlights the importance of selecting appropriate model architectures and validation strategies in the task of classifying Indonesian-language text.