Kerangka Kerja Multimodal Berbasis Adaptive Gated Fusion untuk Mengidentifikasi Aktor Berpengaruh dalam Penyebaran Informasi Demam Berdarah Dengue di YouTube

Lestari, Ayu Eka (2026) Kerangka Kerja Multimodal Berbasis Adaptive Gated Fusion untuk Mengidentifikasi Aktor Berpengaruh dalam Penyebaran Informasi Demam Berdarah Dengue di YouTube. Masters thesis, Institut Teknologi Sepuluh Nopember.

This is the latest version of this item.

[thumbnail of 6026242001-Master_Thesis.pdf] Text
6026242001-Master_Thesis.pdf - Accepted Version
Restricted to Repository staff only
Download (6MB) | Request a copy

Abstract

Demam Berdarah Dengue (DBD) masih menjadi masalah kesehatan serius di Indonesia, dan YouTube menjadi salah satu sumber utama informasi kesehatan yang memengaruhi persepsi publik. Identifikasi aktor berpengaruh dalam penyebaran informasi DBD penting untuk komunikasi kesehatan efektif, namun analisis jaringan sosial umumnya mengandalkan sinyal tekstual dan mengabaikan dimensi visual seperti thumbnail video. Penelitian ini mengusulkan kerangka multimodal berbasis Adaptive Gated Fusion (AGF) yang menggabungkan fitur tekstual IndoBERT dan visual ResNet-50 tanpa modalitas audio. Kedua modalitas difusikan adaptif per dimensi, diagregasikan per kanal, untuk membangun jaringan co-audience berbasis kemiripan konten yang dianalisis menggunakan PageRank, sementara sentimen dianalisis menggunakan RoBERTa berbahasa Indonesia. Menggunakan 33.233 komentar berbahasa Indonesia dari 5.365 video pada 1.084 kanal (2021–2025), ablasi stratified 5-fold cross-validation menunjukkan representasi fusi mencapai akurasi 79,52% dan F1 macro 78,42%, mengungguli baseline lainnya. Analisis jaringan mengidentifikasi KOMPASTV, Kata Dokter, dan Halodoc sebagai aktor paling berpengaruh, membuktikan pengaruh ditentukan posisi struktural, bukan volume komentar. Deteksi komunitas mengungkap beberapa ekosistem audiens, sementara sentimen menunjukkan dominasi respons negatif pada mayoritas kanal berpengaruh. Temuan ini mendukung strategi komunikasi kesehatan berbasis data untuk pencegahan DBD.
===============================================================================================================================
Dengue Haemorrhagic Fever (DHF) remains a serious public health issue in Indonesia, and YouTube has become a major source of health information capable of shaping public perception. Identifying influential actors in DHF information dissemination is essential for effective health communication, yet existing social network analysis approaches largely rely on textual signals and overlook the visual dimension, such as video thumbnails. This study proposes a multimodal framework based on Adaptive Gated Fusion (AGF), combining textual features from IndoBERT and visual features from ResNet-50 without the audio modality. Both modalities are adaptively fused per dimension, aggregated at the channel level, and used to construct a content-based co-audience network analysed using PageRank, while public sentiment is analysed using an Indonesian RoBERTa model. Using 33,233 Indonesian comments from 5,365 videos across 1,084 channels (2021–2025), an ablation study with stratified five-fold cross-validation shows the fused representation achieves 79.52% accuracy and 78.42% macro F1-score, outperforming unimodal and other fusion baselines. Network analysis identifies KOMPASTV, Kata Dokter, and Halodoc as the most influential actors, showing that influence is determined by structural position rather than comment volume. Community detection reveals distinct audience ecosystems, while sentiment analysis shows negative responses dominate most influential channels. These findings support data-driven health communication strategies for DHF prevention.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Analisis Jaringan Sosial, Demam Berdarah Dengue, Adaptive Gated Fusion, IndoBERT, ResNet-50, Aktor Berpengaruh ============================================================ Social Network Analysis, Dengue Hemorrhagic Fever, Adaptive Gated Fusion, IndoBERT, ResNet-50, Influential Actors
Subjects: Q Science > Q Science (General)
Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions: Faculty of Information Technology > Information System > 59101-(S2) Master Thesis
Depositing User: Ayu Eka Lestari
Date Deposited: 29 Jul 2026 01:48
Last Modified: 29 Jul 2026 01:48
URI: http://repository.its.ac.id/id/eprint/139344

Available Versions of this Item

Actions (login required)

View Item View Item