An Intelligent Approach of Vulnerability Severity Assignment Based on Event Extraction

Abstract

The rapid increase in vulnerability reports presents significant challenges for manual analysis and risk management. To address this, we propose an innovative approach for vulnerability severity assignment using deep learning techniques focused on event extraction. Specifically, we develop an event-focused convolutional neural network that processes dual inputs—vulnerability descriptions and related events—to classify vulnerabilities into high, medium, or low risk. This framework improves classification accuracy by effectively incorporating relevant event details. Our approach aims to streamline vulnerability management, enabling organizations to prioritize their cybersecurity efforts more efficiently. Evaluations show notable improvements in accuracy, highlighting the potential of deep learning in vulnerability severity assignment.