Dzaki, Rynofaldi Damario and Syahputra, Razky Ageng (2026) Pengembangan Sistem Agentic Ai Untuk Deteksi Disinformasi, Fitnah, Dan Ujaran Kebencian (DFK) Di Ruang Digital Pada Program Ai Talent Factory Kementerian Komunikasi Dan Digital. Project Report. [s.n.]. (Unpublished)
![[thumbnail of 5025231042_5025231056-Project_Report.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5025231042_5025231056-Project_Report.pdf - Accepted Version Restricted to Repository staff only Download (1MB) | Request a copy |
Abstract
Penyebaran disinformasi, fitnah, dan ujaran kebencian (DFK) di ruang digital berlangsung sangat cepat dan masif di berbagai platform media sosial dan portal berita, sehingga pemantauan secara manual dinilai sudah tidak lagi efisien. Kerja praktik ini dilaksanakan pada Program AI Talent Factory (AITF) Kementerian Komunikasi dan Digital, dengan tujuan membangun system Agentic AI end-to-end yang mampu mendeteksi konten DFK berbahasa Indonesia secara otonom. Sistem yang dibangun mencakup lima tahap utama: crawler untuk mengambil data organik dari platform tren dan lima media sosial (X/Twitter, Instagram, TikTok, YouTube, Facebook), pemrosesan media multimodal menggunakan Vision Language Model dan transkripsi audio, verifikasi fakta dengan pendekatan Retrieval-Augmented Generation (RAG) dan teknik HyDE pada basis data vector Qdrant, klasifikasi konten menggunakan model Bahasa besar hasil fine-tuning (KomdigiITS-8B-DFKClassification- Merged), serta ekstraksi kata kunci untuk memperluas pemantauan secara otomatis. Seluruh kemampuan tersebut disajikan melalui Minimum Viable Product (MVP) berupa dashboard interaktif berbasis Next.js yang dilengkapi alur moderasi manual, jejak audit, galeri bukti, dan visualisasi statistik per platform. Hasil pengujian menunjukkan model klasifikasi teks mencapai akurasi 0,9920, klasifikasi gambar 0,9357, dan ekstraksi kata kunci BERTScore F1 0,8562, sementara seluruh 16 skenario fungsional MVP berjalan sesuai harapan. Sistem berhasil di-deploy pada VPS produksi dan membuktikan bahwa pengawasan isu public skala besar dapat dilakukan secara lebih proaktif dan efisien.
=================================================================================================================================
The spread of disinformation, slander, and hate speech (DS&HS) in the digital space occurs very rapidly and massively across various social media platforms and news portals, making manual monitoring no longer efficient. This internship was conducted under the AI Talent Factory (AITF) Program of the Ministry of Communication and Digital, with the objective of building an end-to-end Agentic AI system capable of autonomously detecting Indonesian-language DS&HS content. The developed system comprises five main stages: a crawler to ingest organic data from trending platforms and five social media networks (X/Twitter, Instagram, TikTok, YouTube, Facebook); multimodal media processing using Vision-Language Models and audio transcription; fact-checking via a Retrieval-Augmented Generation (RAG) approach and the HyDE technique on a Qdrant vector database; content classification using a fine-tuned Large Language Model (KomdigiITS-8B-DFKClassification-Merged); and keyword extraction to automatically expand monitoring scope. All of these capabilities are presented through a Minimum Viable Product (MVP) in the form of an interactive Next.js-based dashboard, equipped with a manual moderation workflow, audit trails, an evidence gallery, and per-platform statistical visualizations. Testing results show that the text classification model achieved an accuracy of 0.9920, image classification reached 0.9357, and keyword extraction obtained a BERTScore F1 of 0.8562, while all 16 functional scenarios of the MVP performed as expected. The system was successfully deployed on a production VPS, proving that large-scale public issue surveillance can be conducted more proactively and efficiently.
| Item Type: | Monograph (Project Report) |
|---|---|
| Uncontrolled Keywords: | Agentic AI, Disinformasi Fitnah Kebencian, Retrieval-Augmented Generation, Fine-Tuning LLM, Content Monitoring |
| Subjects: | Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76 Computer software |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis |
| Depositing User: | Rynofaldi Damario Dzaki |
| Date Deposited: | 07 Jul 2026 07:35 |
| Last Modified: | 07 Jul 2026 07:35 |
| URI: | http://repository.its.ac.id/id/eprint/134425 |
Actions (login required)
 |
View Item |