结合图卷积的在线编程系统成绩预测模型

Performance prediction method of programming online judge combined with graph convolution network

针对传统模型未充分利用学生解答记录的问题,提出一种结合图卷积的在线编程系统成绩预测模型。引入难度与熵权优化深度知识追踪模型,构建学生的知识状态向量;基于知识状态与测试成绩构建学生相似度图,通过图卷积模型融合相似度图中的结点特征;利用融合后的结点特征对学生成绩进行预测。实验结果表明,该模型在真实数据集上相比基线模型能够更准确地预测出学生成绩。

Aiming at the problem that the traditional model does not fully use the records submitted by the students,a method that combined graph convolution network was proposed to predict students’achievement in programming online judgment(POJ).The difficulty and entropy weight were introduced to optimize the deep knowledge tracing model,and the student’s knowledge status was built.A student similarity graph was built based on the knowledge status and POJ scores,and the graph convolution network model was used to fuse the characteristics of the nodes in the similarity graph.The fusion node features were used to predict student achievement.Experimental results show that the model can predict student achievement more accurately than the baseline model on the actual dataset.