【摘要】 针对在线学习平台的高辍学率问题，提出了基于图卷积模型的在线学习辍学预测方法，通过分析学习者在不同时间尺度上的行为特征，及时发现学习者的辍学倾向，采取预防或补足措施.首先，将学习平台采集到的行为数据作为时间序列数据，使用ResNet-50作为局部特征提取的卷积神经网络(CNN),构建包含时间信息的特征向量.其后，将多维特征向量作为图卷积网络(GCN)的节点特征，通过两个GCN网络层建立相关特征的内在联系，并通过数据扁平化尽量保留更多的信息.大规模在线开放课程(MOCC)公开数据集上的实验结果表明，所提方法的预测准确度高于其他先进方法.更多还原

【Abstract】 To address the high dropout rates in online learning platforms, a dropout prediction method based on graph convolution models is proposed. This method analyzes the behavioral characteristics of learners at different time scales to detect dropout tendencies in a timely manner and take preventative or remedial measures. First, the behavioral data collected from the learning platform is used as time series data, and a convolutional neural network(CNN) using ResNet-50 as a local feature extractor is used to construct feature vectors containing time information. Subsequently, multidimensional feature vectors are used as node features for graph convolutional networks(GCN), and the intrinsic relationship between relevant features is established through two GCN layers. Data flattening is used to preserve as much information as possible. Experimental results on a large-scale open online course(MOOC) dataset show that the proposed method outperforms other advanced methods in terms of prediction accuracy.更多还原