Sex trafficking impacts 4.8 million people globally and is a $99 billion USD industry that often operates undetected, including in Canada. Technology has become a critical tool for traffickers, enabling recruitment and exploitation while making these crimes harder to trace. However, innovative analytics can uncover hidden patterns, identify victims, and provide much-needed support to those impacted. Our interdisciplinary team of AI and criminology experts is dedicated to developing context-aware, human-centered solutions to tackle this issue responsibly. Through advanced techniques like data mining and anomaly detection, we are working to bring a data-driven approach to the fight against human trafficking in Canada.
Selected Publications
T-NET: Weakly Supervised Graph Learning for Combatting Human Trafficking
Detecting suspicious ads is challenging due to the sensitive, complex, and unlabeled nature of the data. T-Net addresses this as a weakly supervised graph learning task, leveraging domain-specific signals and ad connections. It introduces a synthetic dataset to aid research while safeguarding privacy.
SWEET : Weakly Supervised Person Name Extraction for Fighting Human Trafficking
SWEET is a weak supervision pipeline for extracting person names from noisy escort ads, addressing the challenge of limited labeled data. SWEET combines rule-matching with fine-tuned large language models as weak labels, effectively aggregating them. Additionally, the HTGEN synthetic dataset is released to support further research in this domain.
TrafficVis: Visualizing Organized Activity and Spatio-Temporal Patterns for Detecting and Labeling Human Trafficking
TrafficVis is an interactive interface for detecting and labeling human trafficking (HT) in clusters of escort ads. Developed with domain experts, it uses advanced text clustering algorithms and metadata signals to visualize spatio-temporal and text patterns. TrafficVis enables experts to label clusters as HT or other suspicious activities, facilitating dataset creation for further research. Expert feedback highlights its usability and explainability, critical for supporting criminal investigations.
INFOSHIELD: Generalizable Information-Theoretic Human-Trafficking Detection
INFOSHIELD is a scalable, parameter-free, and interpretable tool for detecting near-duplicate document clusters, with applications in human trafficking detection, spam-bot identification, and plagiarism. It highlights common phrases, detects unique “slots,” and selects representative documents for clear visualization. INFOSHIELD is language-independent, outperforming domain-specific methods across datasets in multiple languages. It can process 4 million documents in just 8 hours on standard hardware.
Core Team Members

Kellin Pelrine
Mila, McGill

Pratheeksha Nair
Mila, McGill

Mark Ezema
Mila, McGill

Reihaneh Rabbany
Mila, McGill