Abstract: The rapid deployment of massive open online courses (MOOCs) has created a surge in the global connectivity among students for educational purposes. With contrast to large scale enrolled students, almost all the courses in MOOCs have a low completion rate. One of the important reasons for those who drop out the course is lack of suitable study buddies to communicate each other. This paper analyzes students’ posted messages and their roles and behaviors in MOOCs discussion forum, and builds students’ behavior feature model and relational networks between students. To estimate student’s relevancy with respected to topic words dictionary which is selected according to course content in advance, this paper proposes a new topic model with fixed topic dictionary, and finally recommends study buddies for student by inferred topic distribution. Experiments on crawled Coursera dataset demonstrate that the topic model can find out those students with more similarity on same topic words, and recommend study buddies. And this will be helpful for students to complete the study process.

Key words: social network analysis, online courses, recommendation system, topic model

摘要： 大规模开放在线课堂的兴起给高等教育的全球化提供了契机。与大规模参与用户形成鲜明对比的是，目前课程普遍存在完成率不足的问题。造成学生退出的原因之一是学习者缺乏合适的学习伙伴相互交流以及时解决学习中遇到的问题。分析了开放课程中论坛用户的身份特征和学生用户在论坛讨论过程中的行为模式，建立了学习者行为特征模型和学生在讨论过程中形成的关系网络。根据课程内容建立关键词词典，并以此为主题词，提出了一种具有固定主题词的主题模型，进而推断关系网络潜在的主题分布，最终根据主题分布结果为学习者推荐学习伙伴。通过分析Coursera课程平台的真实数据，证明了该学习伙伴推荐方法能有效地挖掘出主题相关的学习者，为学习者相互推荐学习伙伴，在一定程度上将有助于提高学习者的积极性。

关键词: 社会网络分析, 在线课堂, 推荐系统, 主题模型