基于代码结构和时间行为特征提取的庞氏合约检测  

Ponzi Contract Detection Based on Code Structure and Temporal Behavior Feature Extraction

作　　者：秦鉴 王永娟[1] 陆思奇 于刚[1] QIN Jian;WANG Yongjuan;LU Siqi;YU Gang(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2025年第1期76-82,共7页Journal of Information Engineering University

基　　金：河南省重大公益专项(201300210200);国家重点研发计划(2023YFB2705000)。

摘　　要：交易数据特征提取依赖人工设计,不能较好地在时间演化中刻画合约的行为意图;字节码特征提取依靠庞氏和非庞氏智能合约操作码频率分布不同,忽略了合约代码在结构上的特征。提出一个能提取代码结构特征和时间行为特征的模型来提高庞氏合约检测性能。首先,在代码结构特征表示模块中,通过解析合约字节码得到合约控制流图,利用图嵌入技术得到代码结构特征;其次,在时间行为特征表示模块中,根据合约交易构造子图序列,并对每个子图进行嵌入表示,再经过长短时记忆网络处理得到时间行为特征;最后,结合代码结构和时间行为特征来识别以太坊庞氏合约。Feature extraction for transaction data relies on manual design,which fails to adequately capture the evolving behavioral intent of contracts.Bytecode feature extraction,which focuses on the differing frequency distributions of opcodes between Ponzi and non-Ponzi smart contracts,overlooks the structural features of contract code.In the proposed model,features of code structure and temporal behavioral features are extracted to enhance the detection performance of Ponzi contracts.Specifically,in the module for code structure feature representation,the contract control flow graph is obtained by parsing the contract bytecode,and the features of the code structure are derived using graph embedding techniques.In the module for time-behavior feature representation,a sequence of subgraphs is constructed based on contract transactions,with each subgraph being embedded and represented,and then processed using by the long and short-term memory network to obtain the time-behavior features.Finally,the code structure and time-behavioral features are combined to identify Ethereum Ponzi contracts.

关 键 词：区块链 以太坊 庞氏骗局 智能合约 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]