基于增強(qiáng)控制流圖與孿生網(wǎng)絡(luò)架構(gòu)的代碼克隆檢測(cè)方法

打開(kāi)文本圖片集

關(guān)鍵詞：控制流圖；孿生網(wǎng)絡(luò)架構(gòu)；代碼表征；語(yǔ)義相似性；克隆檢測(cè)中圖分類號(hào)：TP311 文獻(xiàn)標(biāo)志碼：A 文章編號(hào)：1001-3695（2025）07-028-2132-09doi：10.19734/j. issn.1001-3695.2024.11.0441

Abstract：Toaddresstheissues of mising contextual information and weak semantic learning capabilities inexisting code clone detection methods，this paper proposedamethod basedonanenhancedcontrolflowgraph（ECFG）and twin network architecture.Firstly，itdesignedECFG，whichembeddedcross-nodecorelationedges tostrenghencontextualawareness. Then，itintroducedCGSMN，asemanticmatching modelbasedontwinnetworks.Thismodelintegratedamulti-headatntion mechanism to extractkeyinformationfromthenodes，thenimprovedtherelational graphatentionnetwork tocapture nternode associationsand generate graph feature vectors.Finall，it explored thesemanticrelationships between thesefeature vectorsandcomputedthesemanticsimilarity.Empirical evaluationwasconductedon tworepresentative datasets.Theresults show that，compared to methods such as ASTNN，F(xiàn)A-AST，and DHAST，the F1 -score on the BigCloneBench dataset improves by0.5 to15.5percentage points，andby1.5to16.5percentage pointsonthe GogleCode Jamdataset，demonstrating the effectiveness of the proposed method for semantic clone detection.

Key words：controlflow graph；siamese neural network；coderepresentation；codesemantic similarity；codeclonedetection

0 引言

近年來(lái)，開(kāi)源文化的蓬勃發(fā)展催生了一系列以協(xié)作共享為理念的開(kāi)發(fā)者社群，對(duì)軟件開(kāi)發(fā)模式產(chǎn)生顛覆性改變[1]，軟件代碼的復(fù)制、粘貼和修改等克隆模式成為一種普遍且高效的實(shí)踐。（剩余23693字）