Pemodelan Topik Dan Analisis Sentimen Topic-Wise Program Makanan Bergizi Gratis (MBG) Menggunakan BERTopic Dan Deep Learning

Auriel, Celine (2026) Pemodelan Topik Dan Analisis Sentimen Topic-Wise Program Makanan Bergizi Gratis (MBG) Menggunakan BERTopic Dan Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221004-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (12MB) | Request a copy

Abstract

Program Makan Bergizi Gratis (MBG) merupakan salah satu kebijakan publik yang banyak dibahas di media sosial dan berita daring karena berkaitan dengan isu gizi, pendidikan, anggaran, implementasi, serta keamanan pangan. Penelitian ini bertujuan untuk mengidentifikasi topik utama mengenai MBG, mengevaluasi performa berbagai pendekatan klasifikasi sentimen, serta menganalisis distribusi sentimen pada setiap topik di Twitter/X dan berita. Data yang digunakan terdiri dari 96.587 dokumen Twitter/X dan 5.176 artikel berita. Metode penelitian dilakukan melalui pengumpulan data, pra-pemrosesan data bercabang, pemodelan topik menggunakan BERTopic, pelabelan manual sentimen, augmentasi data berita, pelatihan model machine learning dan deep learning, serta integrasi hasil topic modeling dan sentiment analysis pada level dokumen. Hasil BERTopic menunjukkan bahwa data Twitter/X menghasilkan 10 final topic substantif dan 1 outlier, sedangkan berita menghasilkan 7 final topic substantif dan 1 outlier. Model sentimen terbaik pada Twitter/X adalah Hybrid BiLSTM IndoBERTweet tanpa menghapus stopwords dengan accuracy 0,9282 dan macro f1 0,9283. Pada berita, model terbaik adalah Ensemble XLM-R Sliding + Hybrid BERT-CNN Sliding dengan accuracy 0,8750 dan macro f1 0,8772. Hasil integrasi menunjukkan bahwa Twitter/X didominasi sentimen negatif sebesar 63,74%, terutama pada isu insiden, anggaran, kualitas menu, dan perdebatan kebijakan. Sebaliknya, berita lebih banyak memuat sentimen positif sebesar 39,75% dan cenderung menyoroti implementasi, kelembagaan, kelompok sasaran, serta manfaat program. Dengan demikian, pendekatan topic-wise sentiment dapat memberikan pemetaan opini publik yang lebih rinci dibandingkan analisis sentimen agregat, karena mampu menunjukkan isu yang dibahas sekaligus polaritas respons terhadap setiap isu.
======================================================================================================================================
The Free Nutritious Meal Program (Makan Bergizi Gratis/MBG) is a public policy that has received substantial attention on social media and online news because it relates to nutrition, education, budget allocation, implementation, and food safety issues. This study aims to identify the main topics surrounding MBG, evaluate the performance of several sentiment classification approaches, and analyze sentiment distribution across topics on Twitter/X and online news. The dataset consists of 96,587 Twitter/X documents and 5,176 news articles. The research process includes data collection, branched text preprocessing, topic modeling using BERTopic, manual sentiment annotation, news data augmentation, machine learning and deep learning model training, and document-level integration of topic modeling and sentiment analysis results. The BERTopic results show that Twitter/X data produced 10 substantive final topics and 1 outlier topic, while news data produced 7 substantive final topics and 1 outlier topic. The best sentiment model for Twitter/X was Hybrid BiLSTM IndoBERTweet without stopwords removal, achieving an accuracy of 0.9282 and a macro f1-score of 0.9283. For news data, the best model was an Ensemble of XLM-R Sliding and Hybrid BERT-CNN Sliding, achieving an accuracy of 0.8750 and a macro f1-score of 0.8772. The integrated analysis shows that Twitter/X was dominated by negative sentiment at 63.74%, mainly related to incidents, budget issues, menu quality, and policy debates. In contrast, news coverage contained a higher proportion of positive sentiment at 39.75% and tended to emphasize implementation, institutional involvement, target groups, and program benefits. Therefore, the topic-wise sentiment Approach provides a more detailed mapping of public opinion than aggregate sentiment analysis, as it captures both the issues being discussed and the sentiment polarity toward each issue.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Analisis Sentimen, BERTopic, Kebijakan Publik, MBG, Topic-wise Sentiment, Public Policy, Sentiment Analysis
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Celine Auriel
Date Deposited:	24 Jul 2026 08:03
Last Modified:	24 Jul 2026 08:03
URI:	http://repository.its.ac.id/id/eprint/137300

Actions (login required)

View Item