ANALISIS MULTIKELAS EMOSI PUBLIK PADA PERIODE AWAL IMPLEMENTASI PROGRAM MAKAN BERGIZI GRATIS MENGGUNAKAN METODE FINE-TUNING INDOBERT PADA APLIKASI X

QAYLLA ANASTASYA BAHAR, . (2026) ANALISIS MULTIKELAS EMOSI PUBLIK PADA PERIODE AWAL IMPLEMENTASI PROGRAM MAKAN BERGIZI GRATIS MENGGUNAKAN METODE FINE-TUNING INDOBERT PADA APLIKASI X. Sarjana thesis, UNIVERSITAS NEGERI JAKARTA.

	Text COVER.pdf Download (964kB)
	Text BAB I.pdf Download (141kB)
	Text BAB II.pdf Restricted to Registered users only Download (1MB) | Request a copy
	Text BAB III.pdf Restricted to Registered users only Download (412kB) | Request a copy
	Text BAB IV.pdf Restricted to Registered users only Download (1MB) | Request a copy
	Text BAB V.pdf Restricted to Registered users only Download (252kB) | Request a copy
	Text DAFTAR PUSTAKA.pdf Download (367kB)
	Text LAMPIRAN.pdf Restricted to Registered users only Download (75kB) | Request a copy

Abstract

Program Makan Bergizi Gratis (MBG) merupakan program pemerintah Indonesia untuk meningkatkan gizi masyarakat, namun pelaksanaannya memunculkan berbagai respon emosional publik yang disampaikan melalui media sosial X. Penelitian ini bertujuan menganalisis multikelas emosi publik terhadap Program MBG menggunakan metode Fine-tuning IndoBERT. Data penelitian berupa 11.184 tweet berasal dari repositori GitHub. Setelah filtering, diperoleh 10.796 tweet yang selanjutnya diberi label emosi secara otomatis menggunakan metode pseudo-labeling berbasis Support Vector Machine (SVM). Tahap preprocessing meliputi cleaning, case folding, tokenisasi, special token, attention mask, dan token ID, yang menghasilkan dataset akhir 10.793 tweet. Data dibagi menggunakan metode Stratified Hold-Out (80% data latih dan 20% data uji). Ketidakseimbangan label ditangani menggunakan class weight dan focal loss, serta konfigurasi model ditentukan melalui hyperparameter tuning. Hyperparameter terbaik diperoleh pada learning rate 5×10⁻⁵, batch size 16, epoch 3, dan dropout rate 0,1. Hasil Fine-tuning IndoBERT menunjukkan bahwa pendekatan class weight memberikan performa terbaik dengan accuracy 81,98%, macro precision 71,96%, macro recall 72,97%, macro F1-score 72,39%, dan weighted F1-score 82%. Hasil klasifikasi ini menunjukkan bahwa emosi marah paling dominan, diikuti senang, jijik, terkejut, takut, dan sedih. Metode Fine-tuning IndoBERT menunjukkan kinerja yang baik dalam klasifikasi multikelas emosi publik terhadap Program MBG. ***** The Free Nutritious Meal (MBG) is an Indonesian government program to improve public nutrition. However, its implementation has diverse public emotional responses expressed through X platform. This study aims to analyze multiclass public emotions toward the MBG Program using the Fine-tuning IndoBERT method. The dataset consisted of 11,184 tweets from GitHub repository. After filtering, 10,796 tweets remained and automatically labeled using a Support Vector Machine (SVM)-based pseudo-labeling approach. The preprocessing stage included cleaning, case folding, tokenization, special token, attention mask, and token ID, resulting in final dataset of 10,793 tweets. The data divided using the Stratified Hold-Out method (80% training and 20% testing). Imbalanced data was addressed using class weight and focal loss, while the optimal model determined through hyperparameter tuning. The best hyperparameter consisted of learning rate of 5 × 10⁻⁵, batch size of 16, 3 epochs, and dropout rate of 0.1. The class weight approach achieved the best performance, with accuracy of 81.98%, macro F1-score of 72.39%, and weighted F1-score of 82.00%. The classification result shows that anger is the dominant emotion, followed by joy, disgust, surprise, fear, and sadness, demonstrating the effectiveness of Fine-tuning IndoBERT for multiclass public emotion classification.

Item Type:	Thesis (Sarjana)
Additional Information:	1). Prof. Dr. Ir. Bagus Sumargo, M.Si. ; 2). Faroh Ladayya, M.Si.
Subjects:	Sains > Statistika
Divisions:	FMIPA > S1 Statistika
Depositing User:	Qaylla Anastasya Bahar .
Date Deposited:	24 Aug 2026 01:25
Last Modified:	24 Aug 2026 01:25
URI:	http://repository.unj.ac.id/id/eprint/72965

Actions (login required)

View Item