PI: Christos Faloutsos
University: Carnegie Mellon University

Given on-line escort advertisements, how can we quickly spot the ones that are near-duplicates, and thus are suspicious for organized, human trafficking? How can we quickly summarize our findings, so that law enforcement can easily decide which leads are promising, and which ones are not? We propose TrafficLight, a system to handle both issues. For the first problem, we will use advanced clustering algorithms based on the so-called 'singular value decomposition,' to automatically group near-duplicates. For the second part, we propose to summarize and highlight the similarities, to make it easier for inspection and verification. The CMU team will continue its collaboration with Marinus Analytics, a PA, female-owned company that does pioneering work in human trafficking detection.