Deteksi risiko human Trafficking pada iklan lowongan pekerjaan di Media Sosial menggunakan Fusion OCR-IndoBERT dan Meta-Feature

Nursyifa, Suci Ihtisabi Hida (2026) Deteksi risiko human Trafficking pada iklan lowongan pekerjaan di Media Sosial menggunakan Fusion OCR-IndoBERT dan Meta-Feature. Sarjana thesis, UIN Sunan Gunung Djati Bandung.

Preview	Text 1_cover.pdf Download (342kB) | Preview
Preview	Text 2_abstrak.pdf Download (242kB) | Preview
Preview	Text 3_skbebasplagiarisme.pdf Download (70kB) | Preview
Preview	Text 4_daftarisi.pdf Download (276kB) | Preview
Preview	Text 5_bab1.pdf Download (290kB) | Preview
	Text 6_bab2.pdf Restricted to Registered users only Download (521kB) | Request a copy
	Text 6_bab2.pdf Restricted to Registered users only Download (521kB) | Request a copy
	Text 7_bab3.pdf Restricted to Registered users only Download (1MB) | Request a copy
	Text 8_bab4.pdf Restricted to Registered users only Download (621kB) | Request a copy
	Text 9_bab5.pdf Restricted to Registered users only Download (230kB) | Request a copy
	Text 10_daftarpustaka.pdf Restricted to Registered users only Download (239kB) | Request a copy
	Text 11_lampiran.pdf Restricted to Repository staff only Download (293kB) | Request a copy

Abstract

INDONESIA: Perdagangan manusia atau dikenal dengan human trafficking mengalami peningkatan siginifikan pada Januari hingga Maret tahun 2025 dibandingkan periode sebelumnya. Perkembangan modus pendekatan baru ke ranah digital melalui iklan lowongan kerja palsu di media sosial semakin sulit dideteksi secara manual akibat volume postingan yang besar dan pola penipuan yang terus berkembang. Penelitian ini bertujuan mengembangkan sistem deteksi otomatis berbasis fusion model yang mengintegrasikan IndoBERT dan meta-feature terstruktur untuk mengklasifikasikan postingan lowongan kerja berisiko perdagangan manusia di Facebook. Sistem dibangun menggunakan pipeline ekstraksi teks berbasis EasyOCR, ekstraksi meta-feature domain-specific, dan model Fusion IndoBERT + Meta-Feature. Eksperimen dilakukan pada 324 sampel postingan dengan 27 konfigurasi yang mencakup variasi arsitektur, strategi fine-tuning, dan rasio pembagian data. Model terpilih menghasilkan akurasi 0,8980, precision 0,8333, recall 0,9524, F1-score 0,8889, dan AUC 0,9439, dengan penambahan meta-feature terbukti meningkatkan recall sebesar 4,76 poin persentase dengan performa yang lebih stabil di seluruh metrik dibandingkan model baseline IndoBERT. Evaluasi cross-validation mengonfirmasi kemampuan generalisasi sistem dengan recall rata-rata 0,8155 ± 0,0641 dan AUC 0,8826 ± 0,0432. ENLISH: Human trafficking saw a significant increase from January to March 2025 compared to the previous period. The emergence of new methods in the digital realm, such as fake job postings on social media, has made manual detection increasingly difficult due to the large volume of posts and constantly evolving fraud patterns. This study aims to develop an automated detection system based on a fusion model that integrates IndoBERT and structured meta-features to classify job postings on Facebook that pose a risk of human trafficking. The system was built using a text extraction pipeline based on EasyOCR, domain-specific meta-feature extraction, and the IndoBERT + Meta-Feature fusion model. Experiments were conducted on 324 sample posts using 27 configurations that included variations in architecture, fine-tuning strategies, and data split ratios. The selected model achieved an accuracy of 0.8980, precision of 0.8333, recall of 0.9524, F1-score of 0.8889, and AUC of 0.9439; the addition of meta-features was shown to increase recall by 4.76 percentage points, with more stable performance across all metrics compared to the IndoBERT baseline model. Cross-validation evaluation confirmed the system’s generalization ability with an average recall of 0.8155 ± 0.0641 and an AUC of 0.8826 ± 0.0432.

Actions (login required)

View Item