Words as the Mightiest Weapon

The first step to defending against our enemies is unmasking them, which can often seem like the biggest challenge. But what if they weren’t masked at all?

Textual data, the conversations happening on public and restricted platforms such as social media and the dark net, is showing itself to be the key to uncovering the radicalized and the radicalizing online.

This kind of data is unstructured by nature and requires formatting before it can function as a threat detector.

“To do analytics, you need numbers, attributes and structure to be attached to the data,” Teradata Senior Data Scientist Jacob Scanlon says. “But there’s a ton of free-form text out there, a potential goldmine of information.”

Detecting unknown threats with text analytics works best if it starts with a known threat, Scanlon says. By comparing the language patterns of the known threat to other text data, like that which is available on Twitter en masse, analysts can start to build lists of other potential bad actors.

But these results are still too big for humans to analyze and too vague to be definitive. Scanlon says that by feeding results from an initial comparison through a psycholinguistic filter, algorithms can be trained to automatically identify certain psychological trends within a body of text — trends that, in the right combination, can indicate a propensity for doing harm.

This is the power of applying structure to unstructured information.

“You can train models to recognize and classify certain kinds of behavior,” says Peeter Kivestu, Industry Consultant at Teradata.

Applying a psycholinguistic filter to the existing predictive analysis heavily culls the count of users that have a high likelihood of being threatening. Once the pool becomes small enough, human analysts can enter the equation to sort the legitimate threats from the “talking heads.”

“This is not replacing the people involved in these operations,” Scanlon cautions. “It’s a force multiplier.”

With a network of people who have a high likelihood of recruiting or otherwise radicalizing, analysts can perform graph analytics to pinpoint the centers of the network — those with the most connections, who are likely the most powerful and the most dangerous.

Detecting threats in a network is about finding a source and going from there, Scanlon says. Identify someone first, then find out who’s connected to him or her.

Kivestu says the beauty of analytics is its preventative capability.

“If all you’re doing is chasing after activities that have already occurred, then your impact may be limited,” he says.

If leaders can take warnings from the behavior they observe, use text analytics to help identify likely motivations around certain illegal behavior — whether it’s violent radicalism or fraud — they can continually run and test data and ultimately stop bad action before it occurs.

Read on here for more information about Teradata and text analysis.

This content is made possible by our sponsor. The editorial staff of Nextgov was not involved in its preparation.