SBIR Phase II: Decoding Obfuscated Text to Find Trafficking Victims

Marinus Analytics Llc, Pittsburgh PA

Investigators

Abstract

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase II project is to combat modern day slavery in the United States and Canada. In this project the company will go beyond the public sector, selling its capabilities to the hospitality industry, which shares a role in tackling this problem. Banks also play a part in detecting financial transactions which stem from criminal revenue streams. These new markets expand the revenue opportunity and social impact of the technology. The proposed Phase II research and development will solve huge challenges voiced by the company's law enforcement users. The deployment of these research products through a platform that enables evidence management and collaboration will accelerate the impact by increasing communication among the fragmented law enforcement jurisdictions in the United States. This will enable agencies to conduct more effective investigations, and empower them to take on larger cases involving organized crime across state lines. Finally, human expertise will be developed within the company and through its partnership with Carnegie Mellon University to commercialize advanced computing research for real-world, social impact. These innovations will empower more victim rescues and exploiter prosecutions. The project will create a culture within the company to nurture engineers in social entrepreneurship. This Small Business Innovation Research (SBIR) Phase II project will expand on machine learning technology created in Phase I to deobfuscate escort ads and implement end-to-end innovations for investigations. Each day, there are thousands of online data points related to prostitution. Hidden behind this content are victims of sex trafficking, those forced or coerced into sex work, including juveniles who have not reached the age of consent. Big Data presents the opportunity to seize this information to disrupt traffickers and organized groups who drive the cycle of exploitation. The company's research objectives include maximizing evidence recall using sophisticated crawlers and deobfuscation methods, as well as generating leads using natural language processing and multi-modal machine learning. The project will further develop computer vision capabilities to interpret features of an image and enable visual search for missing victims. It will formalize methods for collecting ground truth, preventing false positives, and diagnosing algorithmic performance relevant to users' needs. Finally, the company will deploy this research into accessible software products that provide real-time, digestible, and actionable information to law enforcement, resulting in the rescue of hundreds, or potentially thousands, of sex trafficking victims.