IMBWatch -- a Spatio-Temporal Graph Neural Network approach to detect Illicit Massage Business

TL;DR

The paper tackles the covert operations of Illicit Massage Businesses (IMBs), which evade traditional reactive enforcement. It introduces IMBWatch, a Spatio-Temporal Graph Neural Network framework that models IMB ecosystems as dynamic heterogeneous graphs ${\mathcal{G}}=\{G_t\}_{t=1}^T$, where nodes represent parlors, phone numbers, addresses, and advertisements, and edges encode co-location, shared contact info, and synchronized advertising, with temporal attention to capture evolving patterns. Key contributions include domain-informed feature design, integration of OSINT sources, and superior detection performance (F1 up to 84.8%) compared with baselines such as GCN, GAT, ST-GCN, and DCRNN, along with improved interpretability for investigators. Experiments on real-world US-city data demonstrate the model's ability to detect coordinated staff mobility, burner phone reuse, and promotional surges, enabling proactive interventions. The approach offers a scalable, data-driven tool that can adapt to other illicit domains and supports practical deployment via open-source tooling and anonymized data for reproducibility.

Abstract

Illicit Massage Businesses (IMBs) are a covert and persistent form of organized exploitation that operate under the facade of legitimate wellness services while facilitating human trafficking, sexual exploitation, and coerced labor. Detecting IMBs is difficult due to encoded digital advertisements, frequent changes in personnel and locations, and the reuse of shared infrastructure such as phone numbers and addresses. Traditional approaches, including community tips and regulatory inspections, are largely reactive and ineffective at revealing the broader operational networks traffickers rely on. To address these challenges, we introduce IMBWatch, a spatio-temporal graph neural network (ST-GNN) framework for large-scale IMB detection. IMBWatch constructs dynamic graphs from open-source intelligence, including scraped online advertisements, business license records, and crowdsourced reviews. Nodes represent heterogeneous entities such as businesses, aliases, phone numbers, and locations, while edges capture spatio-temporal and relational patterns, including co-location, repeated phone usage, and synchronized advertising. The framework combines graph convolutional operations with temporal attention mechanisms to model the evolution of IMB networks over time and space, capturing patterns such as intercity worker movement, burner phone rotation, and coordinated advertising surges. Experiments on real-world datasets from multiple U.S. cities show that IMBWatch outperforms baseline models, achieving higher accuracy and F1 scores. Beyond performance gains, IMBWatch offers improved interpretability, providing actionable insights to support proactive and targeted interventions. The framework is scalable, adaptable to other illicit domains, and released with anonymized data and open-source code to support reproducible research.

Figures (1)

• Figure 1: Spatio-Temporal GNN framework of IMBWatch