【作 者】：
吴林静， 马鑫倩， 刘清堂， 王瑾洁， 高 喻
【关 键 词 】：
慕课； 大数据； 教师干预； 文本分类； 语义挖掘
【栏 目】：
网络教育
【中文摘要】：
针对慕课论坛中讨论主题数量巨大，教师难以及时反馈的现象，文章提出了一种基于大数据技术的慕课论坛教师干预预测方法。该方法根据学习者的干预需求将慕课论坛中的教师干预分为三种类型：内容相关需干预、管理相关需干预和不需干预。在该分类的基础上，提出了基于词类进行语义特征提取和基于课程知识图谱的内容特征提取方法，对讨论帖的文本内容进行表征，并通过机器学习的方法对教师干预类型进行预测。以中国大学慕课网中“数据库系统概论”课程的教师答疑区主题讨论为数据源，对上述方法的有效性进行验证，发现：（1）基于词类和知识图谱的语义表征方法能够对论坛主题的教师干预需求进行预测，准确率可达到75.86%；（2）不同类型的讨论帖具有不同的语义特征，反映出慕课学习中学习者不同的学习需求，需要教师给予及时、个性化的干预与指导。将慕课论坛教师干预的预测结果推送给慕课教师和课程管理人员，可以大大提升教学管理效率和学习者学习体验。
【英文摘要】：
In view of the large number of topics discussed in MOOC forums and the difficulty for teachers to provide timely feedback, this paper proposes a prediction method of teacher intervention in MOOC forums based on big data technology. This method divides teacher intervention in MOOC forums into three types according to learners' intervention needs: content-related interventions, management-related interventions and no interventions. On the basis of this classification, a semantic feature extraction method based on word classes and a content feature extraction method based on the curriculum knowledge graph are proposed to represent the text content of discussion posts and predict the type of teacher intervention by machine learning methods. Taking the topic discussions of "Introduction to Database System" course in MOOC of China as the data source, the validity of the above methods is verified, and it is found that: (1) the semantic representation method based on word classes and knowledge graphs can predict teachers' intervention needs for forum topics with an accuracy of 75.86%. (2) Different types of discussion posts have different semantic features, reflecting the different learning needs of MOOC learners, which requires timely and personalized intervention and guidance from teachers. Pushing the predicted results of teacher intervention in MOOC forums to MOOC teachers and course administrators can greatly improve the efficiency of teaching management and learners' learning experience.