Disclosing Personal Names in Screen Names Predicts Better Final Achievement Levels in Massive Open Online Courses

Yuan Tian,Wan-Ning Huang,Tai Wang,Lishan Zhang

doi:10.1109/access.2021.3069451

Abstract

The anonymity of the Internet used to be considered as an encouraging factor that helped learners engage in online learning. However, academic studies on anonymity have found that its effect on learning is context-dependent or mixed. In this research, we focused on massive open online course (MOOC) learners’ preference for personal name disclosure in their screen names as a predictor of their final achievement levels (FALs) at the end of a course including 2606 active learners. We conducted two studies, one to examine the associations between these two variables and one to demonstrate how such associations can be utilized in MOOC FAL prediction. We found that MOOC learners who included personal names in their MOOC screen names significantly outperformed other learners in their FALs (p < 0.001). We also found that screen name preference improved FAL prediction accuracy utilizing natural language processing and proper machine learning technologies. The error rate was reduced to 4.03% by a random forest algorithm with an appropriate feature combination: the personal name disclosure indicator (PNDI), quiz scores, number of replies, and exam scores. The results are potentially useful for the development of an early intervention to provide different types of help to students who prefer to disclose personal names and those who do not. The practical effects of these interventions will be examined in the future. In addition, whether the course difficulty level or course type affects the associations between personal name disclosure and FAL will also be examined.

Highlights

Massive open online courses (MOOCs) are large-enrollment, Internet-based classes that provide an alternative to oncampus college courses
Rather than focusing on dropout, we focused on predicting final achievement in the course, which is important since being active throughout the course but still failing the course may frustrate MOOC learners
In Multinomial logistic regression (MLR), due to the personal name disclosure indicator (PNDI), the EQP error ratio increased 3.5%, and the EQR error ratio increased by 0.94%

Summary

Introduction

Massive open online courses (MOOCs) are large-enrollment, Internet-based classes that provide an alternative to oncampus college courses. A typical MOOC lasts 8-15 weeks, with new material provided weekly and time-sensitive assignments due throughout the course. Communication among classmates and instructors occurs through messages posted online using a structure similar to that of social media. This format can be beneficial for many students, MOOCs typically have a low completion rate, suggesting that some students may need support to finish the course [1]. If a learner’s eventual need for support can be predicted at an early stage, the MOOC provider and course instructors. Rather than focusing on dropout, we focused on predicting final achievement in the course, which is important since being active throughout the course (suggesting no risk of dropout) but still failing the course may frustrate MOOC learners. Stakeholders in the MOOC industry may wonder if, besides the indicative Internet behaviors that have been examined in previous studies on MOOC dropout, there are any other Internet behaviors that can help predict final achievement level (FAL) earlier

Objectives

Results

Conclusion