This record contains a peer-reviewed conference poster. The abstract was reviewed and accepted by the conference scientific committee. The poster is also available via the conference website. Abstract- Online job advertisements in the unskilled labor market often exploit vulnerable individuals through fraud and coercion, leading to Internet-facilitated labor trafficking (IF-LT), a pervasive human rights abuse requiring innovative detection methods (Volodko et al., 2020). Compared to sex trafficking, IF-LT is underexplored, with few data-driven initiatives to guide law enforcement and policy. This study presents ExploitScan (Exploitation Scanning Framework), a domain-specific in-development NLP system for detecting exploitative job postings. It operationalizes UNODC (2018) indicators and ILO (2009) measures, enriched with linguistic modifiers and negators, to form a lexicon for weak label generation. These labels, combined with heuristics and classical ML models, yield a baseline classifier, while active learning (Ramachandani et al. 2025) further refines performance, producing a label-efficient, interpretable and scalable pipeline. The novelty of ExploitScan lies in the integration of trafficking research with Artificial intelligence, translating UN/ILO indicators into Machine actionable features yielding context aware, multi step and policy aligned design. Our baseline classifier on manually annotated samples is able to label an ad as “exploitative” which otherwise would be missed if relied on a single indicator set. Future work will explore large language models (LLMs) and multilingual extensions for generalizability.
Pradhan et al. (Wed,) studied this question.