Enhancing security information and event management with W2V-BERT-based real-time threat detection

The growing frequency and sophistication of cyber-attacks have revealed significant limitations in traditional detection systems, highlighting the need for advanced Cyber Threat Intelligence (CTI) solutions. While Open-Source Intelligence (OSINT) provides valuable early-warning capabilities, manual analysis of unstructured data remains time-consuming and error-prone. To overcome these challenges, this study presents an AI-driven framework for real-time cyber threat detection on X, leveraging a hybrid feature extraction approach that combines BERT and Word2Vec embeddings. SHAP-based feature selection is applied to enhance feature relevance and remove redundant or noisy information. Classification is performed using a Deep Neural Network (DNN) optimised for sequential data modelling. Experimental results demonstrate the model’s superior performance, achieving 99.12% accuracy on Dataset 1 and 97.79% on Dataset 2, outperforming existing machine learning approaches. These findings underscore the potential of the proposed W2V-BERT-DNN model to improve CTI by enabling timely, robust, and accurate detection of cyber threats from social media sources.