College launches data portal for counter-terrorism research

Can terrorist groups' past behavior help predict what they might do in different situations in the future?

To a large extent, yes, according to researchers at the University of Maryland who have developed a portal that policy analysts and counter-terrorism groups can use to forecast terrorist behavior based on past actions.

Terminator rising: Killer military robot arms race under way

"This is intended as a platform and an environment [that] Department of Defense (DoD) analysts and others involved in counter-terrorism can use as a way to learn how these groups are operating based on real data," said V.S. Subrahmanian, computer science professor and director of the University of Maryland's Institute for Advanced Computer Studies (UMIACS).

The SOMA Terror Organization Portal (STOP) uses publicly available data on more than 110 terror groups from around the world. It uses a real-time data extraction tool called T-REX to scour and extract data from more than 128,000 articles a day on average from 180 news sites in 93 countries. That data is organized into tables with multiple rows and columns. Each row represents a different year, while each column consists of a variable associated with the group, such as an attack it might have carried out, any counter-measures taken against it by a government, or the level of financial support the group gets from supporters. Each variable then gets a numeric code representing its relative importance.

SOMA, or Stochastic Opponent Modeling Agents, then uses the information and to create rules about the various terrorist groups -- and what they're likely to do -- in its database.

For example, data about the Hezbollah group showed that when it was involved in electoral politics, the chances it will attack civilians outside of Lebanon was in the 69% to 87% range, Subrahamanian said. On the other hand, those chances dropped sharply when Hezbollah is not involved in electoral politics.

According to Subrahamanian, SOMA has generated tens of thousands of such rules about the likely behavior of about 30 groups, including organizations such as Hezbollah, Hamas and Hezb-I-Islami in Afghanistan.

A built-in prediction engine allows policy analysts and other users to run various queries and what-if scenarios against the data, Subrahamanian said. "We have algorithms to try and predict what a group will do in an upcoming or hypothetical situation," Subrahamanian said." A policy analyst might say, 'Here's what the situation seems to be heading towards on the ground and I wonder what this group is going to do based on some action we may take'," he said.

Different variables can be used to create multiple what-if scenarios against the same set of data, he said. In addition to performing queries and running a prediction engine, users can also comment on rules that they find especially useful so others can benefit as well, he said.

In tests conducted by the University, SOMA proved to be accurate in predicting an outcome about 90% of the time. For the tests, university researchers considered each group with at least 10 years worth of data in the database. "What we did was to try and see how correctly we could predict events in the 11th year based on the first ten years of data," he said.

For more enterprise computing news, visit Computerworld. Story copyright Computerworld, Inc.