Linguist uses machine learning for countering violent extremism and radicalization online Meet Falah Al Amro

Language and communication have always played an important role in Falah Al Amro’s life. Growing up, his parents were not able to read or write, and they encouraged him to pursue a better future. Amro is a native Arabic speaker who learned English to expand his skillset and achieve the hopes his family had for him. He earned his bachelor's and master's degrees in linguistics and then his doctoral degree in learning technologies design at George Mason University, cementing a lifelong career in language learning.

Linguist uses machine learning for countering violent extremism and radicalization online

Falah Al Amro joined with the Office of the Director of National Intelligence (ODNI) to foster collaboration between linguistics and social media analysis. He hopes natural language processing algorithms can be used to create an early alert system to detect extremist behaviors online. (Photo Credit: Falah Al Amro)

Linguistics offered Amro a look into the minds of other cultures, as language heavily influences how individuals process concepts. He has developed applications to teach English and worked with agencies who recruited him for Arabic. However, Amro felt this was not enough. So, when he learned about how linguistics and technology could come together to potentially better the globe, he applied for the Oak Ridge Institute for Science and Education’s (ORISE) Intelligence Community Postdoctoral Research Fellowship Program (IC Postdoc) with the ODNI.

The IC Postdoc Program offers scientists and engineers from a wide variety of disciplines unique opportunities to conduct research relevant to the Intelligence Community.

“As you can imagine, one of the best ways to relate to people and cultures is to learn their languages,” explained Amro. “I wanted to use my language and subject matter expertise to contribute to the increasing government needs of leveraging languages and artificial intelligence (AI) for open-source intelligence technologies.

“I am excited to contribute to the field of computer and social sciences to examine the ever-evolving strains of extremist behavior on social media, especially in the rise of hate speech, radicalization, harassment and violence among other ills.”

Amro joined his mentor Hemant Purohit, associate professor for the information sciences and technology department at George Mason University, to collaborate on behavioral modeling using semantic analysis techniques within the field of natural language processing. Natural language processing is a branch of AI in computer science that aims to “teach” computers how to analyze and process language in the “natural” way a human can. An example includes teaching a computer how to understand context and tone in nuanced sentences that use sarcasm, intent or stance.

For Amro’s project, he is looking to minimize the threat landscape from evolving on social media; specifically, he is mining human behavior from social media to address social issues. His goal is to try and understand everything that drives human behavior in English web spaces. He hopes the project will eventually extend to languages outside of English, such as Arabic and its sub-dialects.

“My long-term vision for research is to develop a country-based early warning system that could analyze large corpora of content from many corners of the world with many languages,” said Amro.

The project consists of extracting and analyzing multimodal data from social media. Amro feeds a neural network social media data, and then the network makes a prediction about the attributes of content and the content owner. This can be in the form of predicting trustworthiness, intent, behavior, toxicity, hate or other factors. This is called semantic analysis for human behavior modeling​ and is a crucial part of natural language processing in AI.

“What I often find is that our analysis identifies insights long before it makes it to news channels, or usually not shown on TV channels because it is difficult to detect,” explained Amro. “This is very important as it can provide early warning signs for intelligence purposes, especially when we train our machine learning algorithms on non-Western communication styles, such as Arabic.”

Amro presented his research at the South Big Data Hub Social Cybersecurity Working Group. He shared both the challenges and solutions to processing big social data for understanding malicious public behaviors, and he presented an explanatory model for monitoring, forecasting and prevention.

He recommends the IC Postdoc program. “The program widened my research scope and helped me develop new skills. Most importantly, it allowed me to engage with the research community and make connections in the field.”

After his fellowship, Amro would like to become a federal employee. For now, he continues using his bilingual skills in Arabic and English to bring together linguistics and computer science in a way that he never imagined in the beginning.

The Intelligence Community Postdoctoral Research Fellowship Program is funded by the Office of the Director of National Intelligence (ODNI) and managed by the Oak Ridge Institute for Science and Education (ORISE) under an agreement between the IC and the U.S. Department of Energy (DOE). ORISE is managed for DOE by ORAU.