Doctoral Thesis Oral Defense - Catalina Vajiac
April 7, 2026 10:00AM—11:30AM
Location:
In Person
-
ASA Conference Room, Gates Hillman 6115
Speaker:
CATALINA VAJIAC,
Ph.D. Candidate
Computer Science Department
Carnegie Mellon University
https://csd.cs.cmu.edu/people/doctoral-student/catalina-vajiac
Human trafficking (HT) for forced sexual exploitation is incredibly pervasive, affecting an estimated 6.3 million people at any given time. The majority of victims are advertised online, mainly through online escort websites, alongside at-will escorts. Practitioners who want to help these victims, including government organizations, criminologists, social workers, and investigators, often manually scroll through these escort websites to try to find HT leads by looking for known keywords, geographic movement, or other known HT signals indicating a person was advertised against their will. This manual process is inefficient, as it requires lots of time, and ineffective, as traffickers change their patterns and keywords over time to avoid detection. In addition, since the majority of HT cases are part of organized crime groups, practitioners realized a more reliable HT indicator: groups of ads with nearly-identical text that advertise multiple people, signaling larger organized activity than individual escorts would post. These insights can be leveraged to help facilitate lead generation for practitioners, enabling them to act more quickly to get HT victims out of exploitation.
In this thesis, we assist practitioners in identifying potential HT cases by: (1) developing scalable and explainable clustering algorithms based on text for finding and summarizing organized crime groups in escort ad data, and (2) creating intuitive visualization methods for presenting the results of these to practitioners. These visualizations not only help practitioners to better understand potential leads, but they also facilitate label generation so downstream algorithm evaluation can continue even as traffickers change their patterns. In addition, the methods outlined in this thesis have real-world impact; they are currently being integrated by industry practitioners.
Thesis Committee
Christos Faloutsos (Chair)
Rayid Ghani
Adam Perer
Duen-Horng Chau (Georgia Institute of Technology)
For More Information:
matthewstewart@cmu.edu