VulEye: A Novel Graph Neural Network Vulnerability Detection Approach for PHP Application

Following advances in machine learning and deep processing, cyber security experts are committed to creating intelligent approaches for automatically detecting software vulnerabilities. Nowadays, many practices C C++ programs, methods rarely target PHP application. Moreover, of these use LSTM (Long Short-Term Memory) but not GNN (Graph Neural Networks) learn the token dependencies within source code through different transformations. That may lose a lot semantic information terms representation. This article presents novel Graph Network vulnerability detection approach, VulEye, applications. VulEye can assist researchers finding vulnerabilities projects quickly. first constructs PDG (Program Dependence Graph) code, slices with sensitive functions contained into sub-graphs called SDG (Sub-Dependence Graph), then makes model input train which contains three stack units GCN layer, Top-k pooling attention finally uses MLP (Multi-Layer Perceptron) softmax as classifier predict if is vulnerable. We evaluated on test suite Software Assurance Reference Dataset. The experiment reports show that best macro-average F1 score reached 99% binary classification task 95% multi-classes task. achieved result compared existing open-source implements other state-of-art models. also locate precise area flaw, since our closely related key triggering sensitive/sink function.