Cross-site Scripting Threat Intelligence Detection Based on Deep Learning

In an increasingly complex cyber environment, where the role of traditional protection tools is increasingly limited, intelligence is the key point in the battle. Through the information monitoring of Internet social platforms, potential cyberattack threats to enterprises, governments, and other institutions could be analyzed. Twitter, the world’s largest social media platform, spreads news and shares tweets about cybersecurity-related events and technologies daily, with cross-site scripting attacks being one of them. In the status quo, this paper proposes a cross-site scripting threat intelligence detection model based on deep learning, which can detect tweets involving threats related to cross-site scripting attacks. We utilized a variety of word vector extraction tools blended with topic word extraction techniques to construct a word vector matrix with multi-dimensional features. Then, the threat event detection model is trained using a bidirectional recurrent convolutional neural network with a self-attentive mechanism. In the experiment, the accuracy rate of our proposed model exceeds 0.96, and through multiple sets of control experimental data results, it is proved that the structure designed in the model is conducive to improving the performance of the model and that the model is effective in detecting tweets that involve cross-site scripting threats.