.png){kind=icon}
.png){kind=icon}
Return
Line-Level Smart Contract Vulnerability Detection via Semantic-Syntactic Feature Extraction and Global–Local Attention Network
.png){kind=icon}
PRE
AI
H
L
L
Q
W
DOI：10.1109/tnsm.2026.3714059.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Smart contracts, a core component of Ethereum, automate the execution and management of transactions. However, potential vulnerabilities in smart contracts may trigger serious financial losses or even systemic risks. While existing vulnerability detection methods have achieved certain results, they are still deficient in the precision of line-level vulnerability localization and the breadth of supported vulnerability categories. Against this challenge, this paper proposes a line-level vulnerability detection method with semantic-syntactic feature extraction and global-local attention network. The method utilizes pre-trained model and abstract syntax tree to extract semantic-syntactic features of the code lines, and constructs multiple graph structures to realize the global feature representation of the contract. Based on that, the local network is responsible for extracting line-level features, the global network extracts structured global features, and the two are effectively fused through the attention mechanism. Experimental results show that the method achieves high accuracy and good generalization in the detection of multiple vulnerability categories, with an average F1-score of more than 98% for line-level detection and more than 96% for contract-level detection. This study improves the accuracy of line-level vulnerability detection and provides a valuable technical path for smart contract security research.
Keywords:
Blockchain
smart contract
vulnerability detection
vulnerability localization
deep learning
Journal
IF:
5.4
Papers:
509
Citations:
9.2K
Organization
Cited Papers