基于贝叶斯知识跟踪模型的慕课学生评价 - 北京大学网络与信息系统 .pdf

中国科技论文在线 基于贝叶斯知识跟踪模型的慕课学生评价# 王卓，张铭** （北京大学信息科学技术学院，北京 100871 ） 5 摘要：如何让学生能够更加高效地利用慕课 （Massive Open Online Courses ，简称MOOC ） 上的资源历来就是一个重要的问题，实际上，MOOC 平台的高用户流失率正反应了这方面 的不足。如果能够研究出一个适当的学习者模型，用于对学生学习方面进行建模和解释，将 对进一步提高教学水平起到重要的帮助。而这正是现在MOOC 所缺少的。贝叶斯知识跟踪 10 模型（BKT ）早在上个世纪九十年代便已有人提出，当时被应用在智能教育系统（ITS ）上。 这套模型很好地模拟了学生的学习过程，一直沿用至今。本文将贝叶斯知识跟踪模型应用到 Coursera ，详细分析出 MOOC 的数据特点，据此对 BKT 变形，先后提出 Aspect-BKT 和 History-BKT 两套模型，并进行实验检验，根据实验结果分析 BKT 参数含义。实验结果表 明，History-BKT 与前人提出的IDEM-Count 模型有可比的预测效果。 15 关键词：大规模在线公开课程；贝叶斯知识跟踪模型；学生评价；隐马尔科夫模型 中图分类号：TP311 MOOC Student Assessment Based on Bayesian Knowledge Tracing Model 20 WANG zhuo, ZHANG Ming (School of Electronics Engineering and Computer Science, Peking University, Beijing 100871) Abstract: How let people use the resources on the MOOCs (Massive Open Online Course) efficiently becomes an important issue. In fact, a high churn rate on MOOC reactions on the deficiency. If we could come up with an appropriate learner model, aiming to model and explain 25 the process of learning, it will be helpful to teaching and learning. However, this is now what MOOC lacks. We noticed that Bayesian knowledge tracing model proposed early in the nineties had been applied to the intelligent tutoring system. It models the students learning process well, and still in use now. This paper introduces the Bayesian knowledge tracing model to Coursera, and proves the validity of this model on the MOOC. Besides, we analyze the characteristics of the data 30 on MOOC in detail, and propose two novel mode