融合多维度特征的图卷积网络知识追踪模型

打开文本图片集

关键词：知识追踪；图卷积网络；三元交互图；多维度特征融合；LSTM门控机制中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）12-010-3602-09doi：10.19734/j. issn.1001-3695.2025.06.0156

Multi-dimensional features graph convolutional network knowledge tracing

Fu Rui1，LiXiaolan1，Xu Yan1，He Baidan1，ZhangDaqiang²† （1.DeptofIforationCenter（WuhnDistanceEducationCenter），WuhnVocatioalColgofSfare&Engineeing（WunOpenUni versity），Wuhan430205，China；2.SchoolofSoftuare Enginering，Tongji University，Shanghai ，China）

Abstract：KT，asacoreresearchdirectioninpersonalizededucation，aimstopredictstudents’knowledge states bymodeling their historicalanswersequenes.Toaddressthosecriticallimitationsinexisting models—deficientglobalrelationalmodeling， thedisconnectbetweennode-edge features，anddatasparsity，thispaperproposedamulti-dimensionalfeatures graphconvolutional networkknowledgetracing model（MFGKT）.The modelconstructedastudent-question-knowledge ternaryinteraction graphand designeddual graph convolutioalasociation path ofquestion-knowledge-question andquestion-student-questionto capture highordercorrelations.Furthermore，itenhanced featurerepresentationthroughdepfusionofnode-edge multidimensionalfeaturesanddynamicallyupdated knowledge statesusing anLSTMgating mechanism.Experimentsonthree publicdatasetsdemonstratethatMFGKTsignificantlyoutperformsclasicalmainstreammethods inpredictionaccuracy，efectivelyimpro vingthe precisionandgeneralizationcapabilityofknowledgestate tracing，therebyproviding morereliabletechnicalsupportfor personalized education systems.

Key words：knowledge tracing（KT）；graphconvolutional network（GCN）；ternary interaction graph；multi-dimensional feature fusion；LSTM gating mechanism

0 引言

随着教育数字化与人工智能的深度融合，知识追踪（KT）已成为个性化教育的核心技术。（剩余20996字）