A Survey of Smart Contract Vulnerability Detection Based on Tree Representation

Abstract

Smart contracts, characterized by automation, transparency, and decentralization, can establish efficient, secure, and reliable blockchain fund management and data processing mechanisms, playing a crucial supporting role in blockchain application scenarios. However, security issues in smart contracts are increasingly prominent, making vulnerability detection a critical research focus; detection methods based on tree representation have emerged as a promising direction in this field. A comprehensive review of tree representation-based smart contract vulnerability detection methods is provided in this paper, aiming to offer new insights and theoretical support for smart contract security. First, a three-level vulnerability taxonomy covering 18 common smart contract vulnerability types is presented, categorized by the specific stage of the smart contract life cycle where the vulnerability originates: the Solidity code layer, the EVM execution layer, and the blockchain system layer. Second, four traditional detection paradigms—formal verification, symbolic execution, fuzz testing, and static analysis—are characterized by their core mechanisms and compared across multiple performance dimensions. Third, following the PRISMA systematic review guidelines, 23 primary studies on tree representation-based methods are selected and organized under a novel three-category taxonomy: pure AST sequentialization/traversal, tree-structured neural networks, and AST-to-graph transformations. The technical and application-level challenges of these methods are synthesized, and corresponding open research questions and improvement directions are discussed to inform future work.