基于机器阅读理解的水利工程巡检知识抽取

打开文本图片集

中图分类号：TP391.7 文献标识码：A 文章编号：2096-4706（2025）16-0044-07

Knowledge Extraction for Water Conservancy Engineering Inspection Based on Machine Reading Comprehension

ZHAIXiangchao，ZHANGJianhao，HANWenhao （SchoolofInformationEngeering，NorthinaUiversityofWaterRsourcsandElectricower，Zengzou4o46，Cina）

Abstract： The inspection data of water conservancy projects contains rich risk information. Aiming at the problems ofcomplex long entitiesand nested entities inthedata，ahierarchicalontologymodelandamulti-task kowledgeextraction frameworkareconstructedtosolve theproblemofentityandrelationshipextraction.Firstlyaiming atthecomplexityof domaindata，the hierarchicaltrategy isusedtoefectivelysolvethe modeling problemofcomplex inspectiondata.Secondly，a knowledgeextraction modelcombining machinereading comprehensionand multi-task learning is proposed innovatively.The modelincludesentityextractiontaskbasedonquestionanswering，entityrclasificationtaskbasedondescriptiondisrimination andrelationship extraction task.Each task ealizes collborative optimization through shared parameters and jointtraining. Finally，xperimentalverificationsowstattheentityandrelationshipextractionefectoftheproposedmetodissigiantly beter than otherbaseline models，and itcan meet the actualneeds ofknowledge extractionofengineeing inspectiondata.

Keywords： water conservancy project inspection data ; knowledge extraction ; machine reading comprehension ∴ multitask learning

0 引言

水利工程涉及线路长、地质条件复杂，运行过程中积累了大量工程巡检数据。（剩余9408字）