基于认知诊断与XGBoost的学生表现预测研究

针对认知诊断方法未考虑学生的答题共性和矩阵分解方法未考虑学生知识点掌握个性的问题，提出一种结合认知诊断与XGBoost(eXtreme Gradient Boosting)的学生表现预测方法(PRNCD-XGBoost):首先，根据试题中知识点之间的共现关系探索知识点之间的相似性，并结合试题-知识点二分图挖掘试题中各知识点所占权重，从而进行认知诊断；然后，用认知诊断阶段的预测结果对历史得分矩阵进行填充；最后，采用非负矩阵分解方法提取出包含认知诊断因素的学生答题共性特征进行得分预测。并在ASSISTments2009和ASSISTments2017数据集上，将PRNCD-XGBoost方法与PMF、NeuralCD、PR-NCD、NMF-XGBoost、MNMF-XGBoost等方法进行对比实验。实验结果表明：PRNCD-XGBoost方法在学生表现预测方面具有更高的预测精确度。

Prediction of Students' Performance Based on Cognitive Diagnosis and XGBoost

Aiming at the problem that the cognitive diagnosis method does not consider the commonality of students' answers and the matrix factorization method does not consider the individuality of students' knowledge acquisition, a student performance prediction method (PRNCD-XGBoost) combining cognitive diagnosis and XGBoost (eXtreme Gradient Boosting) is proposed. Firstly, explore the similarity between knowledge according to the co-occurrence relationship of knowledge in exercises, and the weight of each knowledge in the exercise is explored in combination with the exercise-knowledge bipartite graph to make cognitive diagnosis. Then, the students' historical score matrix is filled with the results of cognitive diagnosis. Finally, the non-negative matrix factorization method is used to extract the common features of students including cognitive diagnostic factors for score prediction. The PRNCD-XGBoost was compared with PMF, NeuralCD, PR-NCD, NMF-XGBoost, MNMF- XGBoost methods on the datasets of ASSISTments2009 and ASSISTments2017. The experimental results show that the PRNCD-XGBoost method has higher accuracy in predicting students' performance.