Identifying Extremism in Text Using Deep Learning
The threat of terrorism, both foreign and domestic, is a growing threat to the Western world. Counterterrorism has become a prime task of both law enforcement and intelligence agencies as terrorists use digital media to identify potential recruits and indoctrinate the public. Groups such as Sunni extremist groups, Antifacist terrorists, Sovereign Citizens, and White Nationalists have been responsible for a number of terror attacks and are gaining traction with disenfranchised populations. Traditional methods of investigation are too slow to counter this growing threat, as the methods require a significant amount of manual effort to identify new websites and social media accounts associated with the terror groups. In this paper, we present models that identify extremist propaganda in a multilingual setting with a high degree of accuracy. In this paper, I explore the different forms of terrorism and identify common sources of propaganda for each terror group. Leveraging multicultural and religious experts, a dataset was manually compiled for each form of terrorism. To create a benign dataset for binary classification, I utilized datasets of news articles, encyclopedia entries, religious texts, and miscellaneous postings from social media networks. For models, I compare the efficacy of Long Short Term Memory units (LSTMs) and logistic regression. I show that LSTMs are uniquely positioned to perform text classification and justify the use of deep learning for the task. I evaluate the performance of the models by dataset and achieve an 88% average accuracy with an average F1 score of 0.86. Given the strength of the models, I argue that the models can be used in a real-world application to filter large datasets to identify extremism. I speculate that such models provide a potential alternative to traditional methods of identifying, investigating, and dismantling cyber-enabled terror groups.
Information science|Artificial intelligence
Johnston, Andrew Hammond, "Identifying Extremism in Text Using Deep Learning" (2019). ETD Collection for Fordham University. AAI13882023.