Signal 4 New Alarm to Terrorism: Apply NLP Tech to Improve Public Security

The overstretched budget and huge labor shortage of police force are still burdens to the government, especially nowadays fewer and fewer young people devoted to the police career because of the high risk of death and legal issues. When and where to arrange the limited police force and how many labor persons should be allocated to? To get the answers to these questions, we can ask the help from Natural Language Processing(NLP).

Briefly, NLP model is to transform the linguistic words or phrases into a group of vectors which can be recognized by computer. When you input some new data and model will give the output of the core concepts of the input sentences, or called “token” word.

In an application, a research team from Carnegie Mello University uses the social media and other public online news reports between 1993 and 2012 to train the model in NLP, finding the terrorism related key words. They found that the topic trend kept varying from “refugee” to “sept hijackers” and “deportation”. The new model can stretch out the potential event with the present data.

This technology can be useful not only to find the main issue that the security department should focus om, but also to predict the high risk issues with the extreme words on social media. The method is to find the vector matrix of specific words far away from the main vocabulary matrix. For example, the killer Dylann Roof used to have comment on his action on the Internet, if the model notices a cluster of words totally different from the exiting cluster, the model will give the alarm in advance and police department can rearrange in advance to avoid the tragedy.

Still, the application will meet with a lot of problems like what kind of information should be included in and will it offense people’s privacy. Even working with the model, we still need to keep revising it from time to time to keep the accuracy rate. Another important issue is to avoid people who take use of this function to publish fake arguments or to do hacker attacks to mislead the security system.

Dallas Card, Chenhao Tan, Noah. A. Smith. Neural Models for Documents with Metadata. http://aclweb.org/anthology/P18-1189

Mind-reading DNS security analysis offers early warning for APT attacks. https://www.theregister.co.uk/2015/03/06/precog_dns_security/