关键词: 在线教育, 学生表现, 学习材料, 阅读认知诊断, 知识状态

Abstract: Performance prediction for students（i.e., scores on exercises） is a significant research direction in the domain of online education. However, the prediction accuracy of traditional cognitive diagnosis is insufficient, and collaborative filtering is difficult to guarantee the interpretability of prediction results. Besides, most of the current methods make use of students’response logs on exercises when making predictions. Therefore, they cannot predict the performance of students who have no response logs. Actually, students usually read some auxiliary text learning materials before responding to the exercises. Reading cognitive diagnosis means that the contents of learning materials read by students can frequently reflect students’ knowledge state（i.e., proficiency for knowledge points）, which is beneficial to predict their performance. To this end, this paper proposes a knowledge state modeling method based on reading cognitive diagnosis. First, the students’ potential cognitive level for learning materials is quantified by using the contents that students have read learning materials. Second, the difficulty of learning materials is quantified by combining an educational assumption. Third, according to the above two quantitative results, the students’ actual mastery for learning materials is computed through the item response theory. Based on this, the students’ knowledge state is modeled and their performance can be predicted. Consequently, experiments on a real-world dataset prove that the proposed method can guarantee the accuracy and interpretability of the prediction results. Meanwhile, it can also predict the performance of students who have no response logs.

Key words: online education, student performance, learning material, reading cognitive diagnosis, knowledge state