汽车故障知识图谱构建及应用研究.pdfVIP

汽车故障知识图谱构建及应用研究

摘要知识图谱技术对汽车高效的故障诊断具有重要的意义,现有汽车故障知识图谱构建存在着实体识别模型效果不佳、

无法解决嵌套实体等问题。针对上述问题,通过采用全词掩码的预训练语义模型、加入对抗训练和改进嵌套实体识别模型的

方式提高实体识别模型效果,提出了一种改进的嵌套实体识别模型。实验结果表明,所提模型F1值(F)、精确率(P)和召回

1

率(R)相比基线模型分别提高了3.56%、4.08%、3.05%,相比其他模型也有不同程度的提高,验证了所提模型对汽车维修领

域实体识别具有显著效果。同时,基于构建的汽车故障知识图谱,实现了汽车故障知识智能问答原型系统,展示了知识图谱

技术在汽车故障诊断与维护领域的应用前景。

关键词汽车维修;知识图谱;嵌套命名实体识别;预训练模型;对抗训练

ConstructionandApplicationofFailureKnowledge

GraphinAutomobileField

[Abstract]Knowledgegraphtechnologyisofgreatsignificancetotheefficientfaultdiagnosisofautomobiles.Theexistingconstruc-

tionoftheknowledgegraphofautomobilefaultshasproblemssuchasapoorentityrecognitionmodelandtheinabilitytosolvenested

entities.Inordertosolvetheaboveproblems,animprovednestedentityrecognitionmodelwasproposedbyadoptingthepre-training

semanticmodelofwholewordmask,addingadversarialtraining,andimprovingthenestedentityrecognitionmodel.Theexperimental

resultsshowthattheproposedmodelis3.56%,4.08%,and3.05%higherthanthebaselinemodelintermsoftheF1value(F),

1

accuracy(P)andrecall(R),andalsohasdifferentdegreesofimprovementcomparedwithothermodels,whichverifiesthatthepro-

posedmodelhasasignificanteffectonentityrecognitioninthefieldofautomobilemaintenance.Atthesametime,basedonthecon-

structedautomobilefaultknowledgegraph,theintelligentquestionansweringprototypesystemofautomobilefaultknowledgeisreal-

ized,andtheapplicationprospectofknowledgegraphtechnologyinthefieldofautomobilefaultdiagnosisandmaintenanceisshown.

[Keywords]automobilemaintenance;knowledgegraph;nestednamedentityrecognition;pre-traininglanguag