Thorn Uses AWS to Help Law Enforcement Identify Child-Trafficking Victims Faster

A 2012 Thorn survey of child sex-trafficking survivors found that 75 percent had been sold online, most of them through escort websites. One way for law enforcement agents to catch traffickers and rescue kids would be to identify and investigate escort ads that seem to feature minors, but there are many obstacles to doing this.

One of the most significant obstacles is the massive volume of these ads. As Thorn started work on a tool that could help investigators identify children being exploited, more than 150,000 new escort ads were being posted online every day in the United States. Problems of scale like this cry out for automated processes, but another big obstacle is the fact that escort ad data is anything but consistent: traffickers use code words, slang, and reused content to complicate detection by machines.

To help investigators overcome these and other hurdles, Thorn collaborated with AWS and Digital Reasoning, a company specializing in cognitive-computing solutions, to build Spotlight. Spotlight's machine-learning models analyze new escort ads in real time and use intelligent image analysis and natural-language processing (NLP) to flag those that match risk profiles developed in cooperation with law enforcement agencies. Officers can set customized alerts and search Spotlight’s constantly growing database of ads to aid in their investigations.

Spotlight’s analytics pipeline begins with a third-party solution that ingests more than 100,000 new online escort ads each day, storing text and image data as objects in Amazon Simple Storage Service (Amazon S3) buckets. AWS Lambda functions orchestrate Amazon Simple Queueing Service (Amazon SQS) to queue new ads for processing by the app's analytics engine, Digital Reasoning Synthesys. The solution uses Amazon Simple Notification Service (Amazon SNS) to send alerts to law enforcement agents.

Spotlight's machine-learning models are trained and run on Amazon EC2 C5 instances. Powered by 3.0 GHz Intel Xeon Scalable processors, Amazon EC2 C5 instances offer a memory-to-vCPU ratio that is ideal for demanding inference applications. Spotlight, which is offered free of charge to law enforcement and allied organizations, uses Amazon Cognito to provide a secure user directory that can scale to millions of users.

Amazon Rekognition, which uses deep learning to accurately identify objects, people, activities, and events in images and videos stored in Amazon S3, is at the heart of one of Spotlight’s most crucial capabilities. Facial analysis in Amazon Rekognition helps investigators find photos of the same person that have been edited to defeat less intelligent image-search engines.

"We tested multiple facial-recognition services and found Amazon Rekognition was extremely accurate," says Kristin Boorse, the director of product management for Thorn. "It didn't make sense to build our own facial-recognition service when AWS already offers a best-in-breed solution like Amazon Rekognition. The Amazon Rekognition team has been very responsive in helping us address any challenges we encounter."

Since 2016 alone, Spotlight has helped officers in the United States and Canada open more than 21,000 trafficking cases and identify about 18,000 victims, including more than 6,000 children.

"Our law enforcement partners say Spotlight helps officers identify child sex-trafficking ads much faster than the old paper-and-pencil methods," says Brooke Istook, director of strategy and operations at Thorn. "With the flexibility and easy integration of AWS, we can experiment and prototype new products like Spotlight that would otherwise be cost-prohibitive to develop. On AWS, we can also respond to feature requests from officers in weeks or days—not a level of responsiveness public-sector agencies are used to seeing."

Combating online child sex trafficking is a team effort, according to Istook. " All of our technology partners are making crucial contributions to this fight. Spotlight could never have been so successful without substantial investment and support from AWS, or the underlying technologies that Intel has created.”