A novel framework for semantic classification of cyber terrorist communities on Twitter

Abstract

The exponential growth of Online Social Networks (OSNs) such as Twitter intrigues many terrorist groups to flourish their dark activities, target people to follow and sympathize with their activities, share their ideas, recruit new members, raise funds, and radicalize. In this paper, we focused on identifying terrorist Twitter profiles with high cyber social impact based on semantic tweets analysis and mining techniques. Practically, we proposed a novel framework based on Social Networks Analysis (SNA) and Semi-Supervised Machine Learning (SSML) techniques to classify user accounts and terrorist communities through identifying top influencers by sampling their cyber behaviors. To achieve the targeted goal, we extracted required features using the contemporary topic modeling technique, known as BERTopic. In addition, those features fed into various machine learning classifier models, like SVM, Naïve Bayes, and Logistic Regression classifiers, to find the polarity, which will be used in predicting twitter profiles as extremist or non-extremist accounts. Then, the proposed node classification algorithm uses SNA measures and techniques to identify key players within such extremist communities, and standard classification metrics are used to evaluate the obtained results. Experiments show the efficiency of our framework by outperforming various baseline methods in confronting Twitter extremism.