Detecting Text-Based Cybercrimes Using BERT

The expansion of the internet has significantly reshaped social interactions, particularly through social media platforms where individuals frequently share personal information. However, this increased connectivity also heightens the risk of cybercrime, which often goes undetected due to the anonymity afforded by online activities. Cybercriminals exploit digital platforms while blending into society, making early detection essential to mitigate potential threats. To address this challenge, this research introduces a deep learning model utilizing Bidirectional Encoder Representations from Transformers (BERT) and Natural Language Processing (NLP) techniques to predict potential criminal behavior by analyzing textual data. The model is trained on both Arabic and English datasets. The results demonstrate the model's effectiveness, achieving an average F1-score of 0.82 and an accuracy of 84% for the English dataset, and an average F1-score of 0.87 with an accuracy of 87.3% for the Arabic dataset, highlighting its potential in identifying criminal behavior