.png){kind=icon}
.png){kind=icon}
Return
Exploring extended Abstract Syntax Tree encoding for enhancing code vulnerability detection
.png){kind=icon}
PRE
AI
DOI：10.1007/s11219-026-09748-5.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Vulnerabilities are common in software systems and can potentially pose a serious threat to system security. In response to these threats, various vulnerability detection methods have been proposed recently. Among them, deep learning-based methods have received widespread attention from academia and industry due to their superior detection performance. However, existing deep learning-based methods still struggle to balance detection accuracy, effectiveness, and availability. Moreover, the Abstract Syntax Tree (AST) representation widely used in deep learning-based methods typically treats nodes and edges as static structures, which hinders the accurate portrayal of the complex control flows and the nuanced semantic interrelations that are inherent in dynamic source code interactions. To address these challenges, we propose ExASTrVD ( Extended abstract syntax tree based source code vulnerability detector) to shed light on the dynamic structures between nodes and edges for accurate and efficient vulnerability detection. To facilitate the capture of dynamic deep structural information, ExASTrVD initially constructs an extended AST by carefully adding eight edges with additional information to the original AST and presents multi-relational graphs to store these eight edges. The global embedding vector for multi-relational graphs are derived by applying the adjacency matrix and initial node embeddings. Ultimately, this global vector representation undergoes processing and is integrated into the prediction network for vulnerability detection. Experimental results on large-scale vulnerability datasets show that ExASTrVD is superior to the existing baseline models of vulnerability detection, with higher accuracy, precision, recall and F1-measure. The ablation studies also demonstrate the effective of the core design of ExASTrVD.
Keywords:
Vulnerability detection
Extended abstract syntax tree
Graph attention network
Code representation
Journal
S
IF:
2.3
Papers:
30
Citations:
918