Abstract Collaborative discussions should engage all students, not just a few who dominate (“leaders”) while others participate as “followers” (Zhu, 2006). Cunningham (1991) noted that collaborating learners bring, discuss and debate multiple perspectives to develop their own position while acknowledging others' views. Higher levels of knowledge construction emerged when posts stimulated frequent reply by multiple participants (Aviv, Erlich, Ravid, & Geva, 2003) and were strongly content- and task-oriented (Rovai, 2007). So, to help students more actively and productively engage in knowledge-constructing discussions, an instructor needs to detect students' posts that do not stimulate replies, identify content those posts introduce, and guide students to revise posts to encourage peers' responses. However, such monitoring would be very time- and energy-consuming, especially in large-enrolment courses (Hura, 2010). To set a stage for developing a classifier to automate these tasks, we proposed 10 rhetorical moves characteristic of the interactive mode of Chi and Wylie's ICAP framework (2014) and categorized fine-grained content in discussion posts using these moves. We then identified attributes of posts that triggered a greater number of responses. Rhetorical moves of “asking questions,” “requesting justification,” “building-on,” “giving a reason” and “making a claim” triggered more peer responses. Posts with moves of “disagreeing,” “comparing” and “making claims” predicted students' achievement on a test and an argumentative writing task. We propose analytics for learners and instructors about forming and revising posts to promote constructive discussions and subsequent achievements.

中文翻译：

---

细粒度分析学生在线讨论帖

摘要 协作讨论应该让所有学生参与，而不仅仅是少数占主导地位的（“领导者”），而其他人则作为“追随者”参与其中（Zhu，2006）。Cunningham (1991) 指出，合作学习者带来、讨论和辩论多种观点，以发展自己的立场，同时承认他人的观点。当帖子刺激了多个参与者的频繁回复（Aviv、Erlich、Ravid 和 Geva，2003 年）并且具有很强的内容和任务导向性（Rovai，2007 年）时，就会出现更高水平的知识构建。因此，为了帮助学生更积极、更有成效地参与知识建构讨论，教师需要发现学生发表的不刺激回复的帖子，识别这些帖子介绍的内容，并指导学生修改帖子以鼓励同伴的回应。然而，这种监控非常耗时耗力，尤其是在招生人数较多的课程中（Hura，2010）。为了为开发分类器以自动化这些任务奠定基础，我们提出了 Chi 和 Wylie 的 ICAP 框架（2014）交互模式的 10 个修辞动作特征，并使用这些动作对讨论帖子中的细粒度内容进行分类。然后，我们确定了引发更多回复的帖子的属性。“提出问题”、“要求理由”、“在此基础上”、“给出理由”和“提出主张”等修辞手法引发了更多的同行反应。带有“不同意”、“比较”和“提出主张”等动作的帖子预测了学生在考试和议论文写作任务中的成绩。