Abstract:

The traditional methods of predicting students’grades often require manual screening of characteristics or a large amount of prior knowledge and expert knowledge, the Graph-AE model based on deep learning is proposed to predict students’performance, which can automatically extract features without manual intervention and does not require a lot of prior knowledge. Comparing the Graph-AE model with 13 classical recommendation algorithms, the experimental results show that the Graph-AE model is more accurate on the students’ performance data set than the traditional solutions and can better characterize the relevance and difference between students and courses.

Key words: grade prediction, matrix completion, auto-encoder

摘要：

传统对学生成绩进行预测的方案往往需要手动筛选特征或需要大量的先验知识和专家知识。因此提出使用深度学习的基于图自编码器模型（Graph-AE）的学生成绩预测方案，该模型可以不经人工干预自动提取特征，且不需要大量的先验知识。将Graph-AE模型与13种经典推荐算法进行对比，实验结果表明，Graph-AE模型在学生成绩数据集上的效果比传统解决方法准确度更高，能够更好地刻画学生与课程之间的相关性和差异性。

关键词: 成绩预测, 矩阵填充, 自编码器