INS MMBench：评估LVLMs保险绩效的综合基准 INS-MMBench A Comprehensive Benchmark for Evaluating LVLMs’ Performance in Insurance.docx

下载文档

8
0
约6.54万字
约 71页
2024-07-10 发布于北京
举报
版权申诉
保障服务

INS MMBench：评估LVLMs保险绩效的综合基准 INS-MMBench A Comprehensive Benchmark for Evaluating LVLMs’ Performance in Insurance.docx

1、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。。
2、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
3、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。

INS-MMBench:AComprehensiveBenchmarkforEvaluatingLVLMs’PerformanceinInsurance

ChenweiLin

FudanUniversity

arXiv:2406.09105v1[cs.CV]13Jun2024cwlin23@m.fudan.edu

arXiv:2406.09105v1[cs.CV]13Jun2024

XianXu

FudanUniversity

xianxu@fudan.edu.cn

HanjiaLyu

UniversityofRochester

hlyu5@ur.rochester.edu

JieboLuo

UniversityofRochesterjluo@cs.rochester.edu

Abstract

LargeVision-LanguageModels(LVLMs)havedemonstratedoutstandingper-formanceinvariousgeneralmultimodalapplicationssuchasimagerecogni-tionandvisualreasoning,andhavealsoshownpromisingpotentialinspecial-izeddomains.However,theapplicationpotentialofLVLMsintheinsurancedomain—characterizedbyrichapplicationscenariosandabundantmultimodaldata—hasnotbeeneffectivelyexplored.Thereisnosystematicreviewofmulti-modaltasksintheinsurancedomain,norabenchmarkspecificallydesignedtoevaluatethecapabilitiesofLVLMsininsurance.ThisgaphindersthedevelopmentofLVLMswithintheinsurancedomain.Inthispaper,wesystematicallyreviewanddistillmultimodaltasksforfourrepresentativetypesofinsurance:autoinsurance,propertyinsurance,healthinsurance,andagriculturalinsurance.WeproposeINS-MMBench,thefirstcomprehensiveLVLMsbenchmarktailoredfortheinsurancedomain.INS-MMBenchcomprisesatotalof2.2Kthoroughlydesignedmultiple-choicequestions,covering12meta-tasksand22fundamentaltasks.Furthermore,weevaluatemultiplerepresentativeLVLMs,includingclosed-sourcemodelssuchasGPT-4oandopen-sourcemodelslikeBLIP-2.Thisevaluationnotonlyvalidatestheeffectivenessofourbenchmarkbutalsoprovidesanin-depthperformanceanalysisofcurrentLVLMsonvariousmultimodaltasksintheinsurancedomain.WehopethatINS-MMBenchwillfacilitatethefurtherapplicationofLVLMsintheinsurancedomainandinspireinterdisciplinarydevelopment.Ourdatasetand

evaluationcodeareav

INS MMBench：评估LVLMs保险绩效的综合基准 INS-MMBench A Comprehensive Benchmark for Evaluating LVLMs’ Performance in Insurance.docx 原文免费试下载