Prediction of Pre-Radicalism Leading to Hate Speech in Social Media Accounts Using Machine Learning and Intelligence Data Gathering Frameworks

Abstract

Social media presents concerns about extremist content spreading which produces anxiety about radicalization that occurs through online means. Digital platforms provide two advantages to evil organizations by helping them recruit new members and conducting their activities while distributing propaganda. BERT (Bidirectional Encoder Representations from Transformers) within the designed system detects indicators of pre-radicalization to prevent detecting the onset of extremist behavior and hateful speech. The system draws from Twitter data collected by intelligence data gathering frameworks and conducts ongoing training updates to enhance its operational efficiency as well as adaptability capabilities. Text normalization begins a preprocessing sequence which is followed by tokenization operations before companies perform feature extraction. Evaluations of the model happen through precision and recall measurements as well as accuracy assessment followed by F1-score evaluation. The identification of radicalization warning signs by separate deep learning algorithms proves more effective than traditional systems at early stages of development. AI systems for radicalisation detection need continuous training which results in improved detection capabilities and allows their application in both counterterrorism operations and political extremism support programs and online hate speech monitoring.