大语言模型在软件测试方面的研究综述

打开文本图片集

Abstract：WiththerapiddevelopmentofLargeLanguageModel（LLMs）technology，itsapplicationpotentialinthefieldof softwaretestingisincreasinglyprominent.ThispaperexpoundsthebackgroundandsignificanceofLLMspromotingtheinovation ofthetestingparadigm.StartingfromtheTransformerarchitecture，thispaperdeplyanalyzeshowitspre-trainingandfinetuningtrainingmechanismendowsthemodelwiththeabilityofgeneralsemanticunderstanding，andexploresthesuportingrole oflongcontextmodelingandmultimodalexpansionforcomplextasks.ThepracticalaplicationsofLLMsintestcasegeneration， defectdetectionandanalysis，testscriptautomation，testresultanalysisandreporting，andtestcoverageoptimization weremainly discused.Acomprehensivereviewof thechalengesandprospectsofLLMswasconducted：Issuessuchasmodeldata limitations，technicalobstacles，andengineringimplementationobstaclesstillneedtobeovercome.Inresponsetothese problems，future research directions were proposed.

Keywords：Large Language Models;softwaretesting;testcasegeneration;defectdetectionandanalysis;Transformer architecture

随着软件系统复杂性的指数级增长，软件测试作为保障软件质量的核心环节，已成为现代软件开发流程中不可或缺的支柱。（剩余9834字）