October 16, 2017
For all its benefits, technology has spawned new ways to conduct and coordinate criminal activity. The dark web, where users can remain anonymous and untraceable, is the prime example. Naturally, law enforcement agencies have taken a great interest in finding ways to gather and sort out information from these channels to stop criminal and terrorist networks. Fortunately, technology can also help thwart this illegal activity.
Using computer science, Professor Sucheta Soundarajan and Ph.D. student Pivithuru Wijegunawardana are developing a way to help authorities identify people of interest based on their personal connections. The idea is that if they are able to identify a person that has been involved in a given activity, they may be able identify others.
In their research, “Seeing Red: Locating People of Interest in Networks,” the researchers have modeled a multilayered terrorist network. Each layer is defined by different kinds of personal relationships, including belonging to the same organization, attending the same schools or training together, kinship, and so on.
The researchers start with data from a single person—someone that has been officially classified as a terrorist by authorities. Then, using their learning-based algorithm called RedLearn, they branch out from that person to predict which of his or her associates are likely to be involved in similar activities. By translating human relationships, behaviors, and interactions into cold, hard data, they can be fed through RedLearn and be used to accurately reveal as many people of interest as possible.
“The is no automated algorithm that can be trusted to classify people as terrorists or criminals. This task is still best left to the living, breathing experts in law enforcement. However, our algorithm can narrow down the data from places like the dark web to recommend where investigators should consider looking deeper,” said Soundarajan.
Soundarajan and Wijegunawardana’s high-tech solution takes this practice to a higher level of accuracy and reach than previous methods. What makes RedLearn truly special is the novel way it enhances its effectiveness by going beyond simple associations—it takes dishonestly into consideration.
In the context of identifying potential criminals or terrorists, it is fair to assume that individuals may provide false information to shield themselves and others. This means the collected data may be deceptive. RedLearn takes this probability into account and instead of throwing investigators off, lies can actually help them focus in on a person of interest.
And, it’s remarkably effective, beating the next best strategy by up to 340 percent.
“To our algorithm, criminal vs. non-criminal or true vs. false is just 1 vs. 0 or red vs. blue. It’s still up to human analysts make decisions about which data should be further explored,” said Wijegunawardana. “Our algorithm helps point them in the right direction.”
Given their solutions’ neutral position, the researchers see that it can be applied to many scenarios beyond law enforcement. It could be used to identify any subpopulation within a larger population, such as students in danger of failing in within a University.
Soundarajan and Wijegunawardana are pleased that this work can be used to help stop those that would seek to harm others, and are proud to have developed a solution that could easily be put to use in other, less nefarious communities.