Deteksi Potensi Gangguan Depresi Pada Media Sosial X Berbasis User-Level Analysis Menggunakan Arsitektur Multi-Modal Transformers

Taratugang, Mohammad Arkananta Radithya (2026) Deteksi Potensi Gangguan Depresi Pada Media Sosial X Berbasis User-Level Analysis Menggunakan Arsitektur Multi-Modal Transformers. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027221003-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Kesehatan mental merupakan isu krusial yang pengungkapannya kini banyak beralih ke media sosial "X" melalui unggahan teks dan gambar. Mengatasi keterbatasan sistem deteksi saat ini yang mayoritas berfokus pada analisis per-cuitan (post-level), penelitian ini mengembangkan sistem identifikasi potensi gangguan depresi berbasis analisis tingkat pengguna (user-level) guna mendeteksi konsistensi gejala klinis sesuai kriteria DSM-5. Metode yang digunakan adalah arsitektur Multimodal Transformer melalui fine-tuning model IndoBERTweet untuk ekstraksi fitur teks dan Vision Transformer (ViT) untuk ekstraksi fitur gambar melalui mekanisme Late Fusion. Model dilatih menggunakan dataset riwayat linimasa pengguna dan didukung teknik Explainable AI (XAI) melalui visualisasi SHAP Partition Explainer guna menjamin transparansi klinis, kemudian diimplementasikan ke dalam browser extension dan dashboard pemantauan untuk memindai 50-100 unggahan terakhir secara otomatis. Hasil pengujian membuktikan bahwa arsitektur fusi Multimodal ViT berhasil mencapai akurasi sebesar 82,10% dan F1-Score 79,10%, serta secara signifikan mampu menekan angka False Negative dibandingkan arsitektur teks tunggal. Validasi kualitatif oleh pakar psikologi klinis mengonfirmasi bahwa fungsionalitas sistem telah selaras dengan nalar medis, disertai penyesuaian terminologi etis keluaran sistem menjadi "Indikasi Stres" guna mencegah self-diagnosis. Lebih lanjut, pengujian pengguna akhir (End-User Testing) memperoleh skor System Usability Scale (SUS) sebesar 80, yang menegaskan bahwa sistem ini dapat diterima (Acceptable), mudah dioperasikan, dan layak dijadikan sebagai instrumen skrining awal (early screening) yang proaktif dan bertanggung jawab.
=============================================================================================================================================
Mental health is a crucial issue whose expression has increasingly shifted to the "X" social media platform through text and image posts. To overcome the limitations of current detection systems that predominantly focus on post-level analysis, this research develops a potential depressive disorder identification system based on user-level analysis to detect the consistency of clinical symptoms according to DSM-5 criteria. The methodology involves a Multimodal Transformer architecture through the fine-tuning of the IndoBERTweet model for text feature extraction and Vision Transformer (ViT) for image feature extraction via a Late Fusion mechanism. The model was trained using a user timeline history dataset and supported by Explainable AI (XAI) techniques through SHAP Partition Explainer visualizations to ensure clinical transparency, before being implemented into a browser extension and a monitoring dashboard to automatically scan the last 50–100 posts. Evaluation results prove that the Multimodal ViT fusion architecture successfully achieved an accuracy of 82.10% and an F1-Score of 79.10%, significantly reducing the False Negative rate compared to the text-only architecture. Qualitative validation by a clinical psychologist confirmed that the system's functionality aligns with medical reasoning, accompanied by an ethical terminology adjustment of the system's output to "Stress Indication" to prevent self-diagnosis. Furthermore, End-User Testing obtained a System Usability Scale (SUS) score of 80, confirming that the system is acceptable, user-friendly, and viable as a proactive and responsible early screening instrument.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Depresi, IndoBERTweet, Media Sosial X, Explainable AI, Vision Transformer, DSM-5, Browser Extension, Depression, Social Media X
Subjects:	T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Mohammad Arkananta Radithya Taratugang
Date Deposited:	29 Jul 2026 02:04
Last Modified:	29 Jul 2026 02:04
URI:	http://repository.its.ac.id/id/eprint/139430

Actions (login required)

View Item