Online labor trafficking and machine learning: A scoping review and research agenda

Labor trafficking increasingly intersects with online environments, yet there are limited computational detection approaches. To examine existing methods and their potential applications, this article reviews literature that applies or advocates analytics and machine learning (ML) for detecting labor trafficking (LT). It synthesizes methodologies across two contexts: (a) internet-facilitated LT (IF-LT), where digital platforms enable recruitment and coordination of trafficking, and (b) online-accessible LT (OA-LT), where offline signals of trafficking can be retrieved from online sources. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines, six bibliographic databases were searched between January 2002 and September 2025 and complemented with additional journals, backward and snowball searches, and expert recommendations. Of 2,151 records, we assessed 51 articles for full-text and 21 were mapped and qualitatively synthesized (14 LT-focused and 7 methodologically transferable). IF-LT research clusters around online recruitment, recruitment fraud, illicit massage businesses (IMBs), while OA-LT examines supply chains, child labor, violence indicators on social media, modern slavery disclosures and high-risk sectors such as the fishing industry. Common approaches include case studies with explainable analytics, interpretable classifiers, topic modeling and advanced natural language processing (NLP) methods like sentiment analysis, transformers and large language models; positive-unlabeled learning, probabilistic modeling, and spatial analysis. Despite attention to online LT, the review reveals a gap in quantitative applications of state-of-the-art ML to online content for LT detection. Interdisciplinary and transferable techniques developed for detecting online sex trafficking (ST) offer an adaptable pathway towards automation and law enforcement applications. Beyond methodological gaps, the review emphasizes responsible and context-aware AI, including weighted indicator modeling, remedies for class imbalance, multimodal data integration, label-efficient learning with expert annotations, and clear diagnostic reporting to support accountable deployment.