Analisis Sentimen Masyarakat Terhadap Kebijakan Efisiensi Anggaran Pemerintah Pada Platform X Menggunakan Metode IndoBERTweet

Khairani, Zahara (2026) Analisis Sentimen Masyarakat Terhadap Kebijakan Efisiensi Anggaran Pemerintah Pada Platform X Menggunakan Metode IndoBERTweet. Skripsi thesis, Universitas Islam Negeri Imam Bonjol Padang.

	Text (cover) ZAHARA KHAIRANI_2217020057_cover-daftar isi.pdf - Published Version Download (1MB)
	Text (File BAB I) Bab 1.pdf - Published Version Download (361kB)
	Text (File BAB 3) Bab 3.pdf - Published Version Download (422kB)
	Text (File BAB 5) Bab 5.pdf - Published Version Download (366kB)
	Text (fulltext) ZAHARA KHAIRANI_2217020057_Fulltext.pdf - Published Version Restricted to Repository staff only Download (3MB)

Abstract

The implementation of the government budget efficiency policy in 2025 has generated various public responses on Platform X (formerly Twitter). The large number of public opinions expressed through textual data makes manual identification of sentiment tendencies difficult to perform effectively. This research aims to analyze public sentiment toward the government budget efficiency policy on Platform X using the IndoBERTweet model based on the Transformer architecture. The dataset used in this research consists of Indonesian-language posts collected from Platform X during January to December 2025. After undergoing data selection and manual labeling processes, a total of 3,228 data were obtained and classified into three sentiment categories: negative, neutral, and positive. This research consists of several stages, including data collection, preprocessing consisting of cleaning, case folding, and normalization, handling class imbalance using Random Oversampling (ROS) applied only to training data, fine-tuning the IndoBERTweet model, and evaluating model performance using confusion matrix and classification metrics, including accuracy, precision, recall, and F1-score. The model evaluation was conducted using three evaluation schemes, namely 80:20 data splitting, 70:30 data splitting, and 5-Fold Cross Validation. The evaluation results show that the IndoBERTweet model achieved stable performance across all evaluation schemes. The 80:20 and 70:30 data splitting schemes obtained an accuracy of 82%, while the 5-Fold Cross Validation scheme achieved an accuracy of 81%. The Macro F1-score values obtained were 0.76 for the 80:20 data splitting scheme, 0.76 for the 70:30 data splitting scheme, and 0.76 for the 5-Fold Cross Validation scheme. These results indicate that the IndoBERTweet model provides consistent performance in classifying public opinions regarding the government budget efficiency policy despite the imbalanced distribution of sentiment classes.

Actions (login required)

View Item