Virginia Tech® home

Countering Human Smuggling by Identifying Deceptive Advertisements in Social Media

Researchers will develop artificial intelligence-powered detection mechanisms to monitor and detect mis/disinformation in social media ads targeting migrants, and flag them for manual inspection and investigation.

Project funded by: CCI Hub

Project Investigators

Principal Investigator (PI): Faryaneh Poursardar, lecturer, Old Dominion University Computer Science Department

Co-PI: Vikas Ashok, assistant professor, Old Dominion University Computer Science Department

Co-PI: Erika Frydenlund, research assistant professor, Old Dominion University's Virginia Modeling Analysis and Simulation Center (VMASC)

Rationale and Background

Around the world, migrants plan journeys to other countries based on information found on social media.

Intentional misinformation and/or disinformation can play a large part in humanitarian crises involving human smuggling and trafficking. 

Human smugglers (also known as coyotes on the U.S. southern border) often advertise on social media, sometimes presenting posts that appear to represent legitimate travel agencies or companies, with subtle visual or textual cues indicating that services include assistance to illegally cross a border. 

Migrants can fall victim to targeted mis/disinformation from smugglers. With the shifting nature of profiles and posts on social media, there will continue to be outlets for smugglers to connect with migrants desperate to cross. 

Methodology

  • Develop a deceptive advertisement dataset: Create a custom dataset consisting of both diverse deceptive advertisement posts and legitimate travel-agency advertisements in different languages from online sources and human collaborators.
  • Develop accurate machine learning-based detection algorithms: Leverage that dataset to train, validate, and test machine-learning models that can accurately distinguish between legitimate and deceptive advertisements.

Projected Outcomes

  • A trained transformer-based classification model to distinguish between deceptive and legitimate advertisements.
  • Code for training the model.
  • Evaluation results of the model.
  • Publications disseminating the findings.