Development of a student engagement score for online undergraduate engineering courses using learning management system interaction data
Development of a student engagement score for online undergraduate engineering courses using learning management system interaction data
复制标题
DOI：
10.1002/cae.22479
复制
发表时间：
2021-11
影响因子：
2.9
通讯作者：
Javeed Kittur;J. Bekki;Samantha R. Brunhaver
中科院分区：
文献类型：
--
作者：
Javeed Kittur;J. Bekki;Samantha R. Brunhaver
尽管研究人员一致认为，学生对在线课程的参与度是与课程相关活动的时间相关的，但对于量化这一结构的最佳方法却几乎没有达成共识。本研究介绍了一种通过本科工程学生与在线课程学习管理系统 (LMS) 的互动来衡量他们参与在线课程的方法。来自三个完全在线的本科工程学位课程提供的 81 门课程的数据总共生成了 3848 个独特的学生课程组合（大约 270 万行 LMS 交互数据），我们应用了五步流程来计算代表学生 LMS 参与度的单个分数。首先，我们将学生的 LMS 交互数据转换为一组自然特征，代表他们每 3 天在各种课程元素（例如测验、作业、论坛等）上花费的时间，以及这些时间在课程期间如何变化。然后，我们使用自然特征导出 216 个相关特征，描述同一课程中学生之间典型交互模式的偏差。接下来，我们对数据集的训练部分进行关联规则挖掘，以生成分别描述完成课程的学生（完成者）和选择提前退学的学生（离开者）行为的规则。生成的规则应用于数据集测试部分的学生，以计算完成者和离开者满足的独特规则的百分比。最后，发现每个学生满足的完成者和离开者规则百分比之间的数学差异是衡量学生参与度的最佳衡量标准。
Although researchers agree that student engagement in online courses is a function of time dedicated to course‐related activities, there is little consensus about the best way to quantify the construct. This study introduces a measure for undergraduate engineering students' engagement in online courses using their interactions with their online course learning management system (LMS). Data from 81 courses offered by three fully online, undergraduate engineering degree programs generated a total of 3848 unique student–course combinations (approximately 2.7 million rows of LMS interaction data), to which we applied a five‐step process to calculate a single score representing student LMS engagement. First, we converted the students' LMS interaction data into a set of natural features representing the time they spent per 3‐day period on various course elements, such as quizzes, assignments, discussion forums, and so forth, and how these times changed across the duration of the course. We then used the natural features to derive 216 relative features describing deviations from typical interaction patterns among students in the same course. Next, we conducted association rule mining on a training portion of the data set to generate rules separately describing the behavior of students who completed the course (completers) and those who chose to drop early (leavers). The rules generated were applied to students from the testing portion of the data set to compute the percentage of unique rules met by completers and leavers. Finally, the mathematical difference between the percentages of completer and leaver rules met by each student was found to be the best measure of student engagement.