在线学习群体检测是在新一轮科技革命赋能教育创新变革背景下,依据学习者个性化特征优化教育资源分层配置的关键途径.现有学习趣缘社群在线学习群体的检测主要依赖学习者的直接行为记录和互动指标,较少关注学习者潜在的社交参与水平和社群结构.为营造数智环境下学习者画像决策辅助全民自主学习的文化氛围,本文提出一种社交参与视角下超图增强的学习趣缘社群群体检测方法.首先,从影响用户社交参与的维度出发,构建能够体现学习者社交参与水平的特征集.其次,提出超图卷积网络(hypergraph convolutional network,HyperGCN)增强的图聚类算法HG-SDCN(structural deep clustering network based on HyperGCN),解决了利用二分图检测在线学习群体时无法有效捕捉学习者多元交互关系和高阶结构的问题.最后,从真实学习趣缘社群收集数据,验证本文提出方法的检测效果.与基线相比,本文方法在Acc(accuracy)、F1、NMI(normalized mutual information)和ARI(adjusted Rand index)等评价指标上分别提升了16.16、9.77、16.01和22.14个百分点.上述结果不仅证明了HyperGCN在捕捉学习者高阶结构实现在线学习群体检测任务中的有效性,还为未来从社交参与维度制定调整个性化教育资源配置策略提供了方法和理论支撑.
Group Detection in Interest-Based Learning Communities Enhanced by Hypergraph from a Social Engagement Perspective
Online learning group detection in the context of the new technological revolution empowering educational in-novation,is a key approach for optimizing the stratified allocation of educational resources based on the personalized char-acteristics of learners.Existing detection methods for interest-based learning community online learning groups primarily rely on direct behavioral data and interaction metrics of learners,focusing less on the potential level of social engagement and community structure.To cultivate a culture of autonomous learning enhanced by learner profiles in a smart digital envi-ronment,we propose a group-detection method for interest-based learning communities enhanced by a hypergraph from a social engagement perspective.Initially,a feature set representing the learner's level of social engagement is constructed based on factors influencing users'social engagement.Subsequently,a hypergraph convolutional network(HyperGCN)-enhanced graph clustering algorithm is proposed to overcome the issue of ineffective capture of multivariate interactions and higher-order structures of learner groups previously encountered with bipartite graph detection.Data were collected from a real-life interest-based learning community to validate the effectiveness of the proposed method.Compared with the baseline,the proposed method achieved improvements of 16.16,9.77,16.01,and 22.14 percentage points in accuracy(Acc),F1,normalized mutual information(NMI),and adjusted Rand index(ARI),respectively.These results not only prove the effectiveness of HyperGCN in capturing the structure of learner groups for online learning group detection tasks but also provide methodological and theoretical support for formulating and adjusting personalized education configuration strategies from the perspective of social engagement.
group detectionhigher-order structuresocial engagementhypergraph convolutional networkinterest-based learning communitypersonalized learning
NETL
NSTL
万方数据
