Investigating Students’ Learning through Learner roles and Linguistic Expressions in Massive Open Online Courses

Abstract

In recent years, there has been increasing interest in online learning environments, notably in Massive Open Online Courses (MOOCs). Due to such interest, predictions and education data mining have rapidly gained prominence in education studies over the past decade. As many of the MOOCs are freely available for students, they draw the interest of thousands of learners. However, assessing the success rate of student learning through online platforms has become difficult to quantify as students enroll for varying purposes, such as browsing the course content or enrolling into similar courses to find the best fit for their needs. Knowing that students may enroll in courses for other purposes, research studies need to explore diverse perspectives of learning success beyond completion. The massive amount of student data available in MOOC platforms enables researchers to gain valuable insights into students’ learning behaviours, enabling analysis of aspects such as performance predictions and cognitive engagement. Using discourse analysis, it is possible to investigate the learner-generated discourse from discussion forums to understand students’ learning in many ways so as to identify information-seeking learners, to identify linguistic behaviours of students working at different grades and to understand how well a student has understood course content; in particular, their issues and knowledge on a specific content across the course. This thesis explores the concepts of ‘student roles’ and ‘linguistic expressions’, which can be extracted from discussion forums to investigate students’ learning across AdelaideX MOOCs. Using the grounded theory approach, the thesis identifies student roles in the study data as ‘information-seeker’, ‘information-giver’ and ‘other’. By identifying these roles, it became possible to determine the student roles in discussion forums where a lack of peer interactions is observed. This thesis aims to categorise these roles solely based on discourse analysis by leaving the contextual (e.g., previous post) and community-related features (e.g., views) behind. The existing literature has also identified a number of important community-related structural features, such as structural position in thread and number of votes; however, a challenge is that they are not feasible to incorporate in real-time predictions as they are dynamic and change throughout the course. Furthermore, waiting for such community-related features requires more time and effort to predict these student roles in discussion forum posts. This thesis bridges this gap by predicting the student roles based solely on analysing the linguistic expressions (e.g., word count, cognitive level and analytical thinking) extracted from the content of posts. Moreover, the thesis also identifies the learner topics that have been discussed in the forums and measures the correlation between learner clusters and topics. Exploring the correlations, such as topic contributions with course grades, topic contributions with student roles and so forth, helps to identify how different groups of learners are contributing in discussion forums. Going beyond student roles, this thesis presents a linguistic-based rule set to identify at-risk learners based on the linguistic contribution they have made in discussion forums which may support educators and researchers to find the associations between usergenerated content and final course grades. These rule sets are generated by considering the learners’ optional participation, which can be seen in many MOOCs. Lastly, this thesis investigates the linguistic expressions of pass and fail grade learners with time for two different discussion forum components, namely comment threads and comments. Furthermore, ‘Linguistic Profiles’ for two different learner grades were proposed that can be used as a template to distinguish their linguistic behaviours. Based on these investigations, the thesis provides empirical evidence of where roles exhibited by a student and their language use in discussion forums can help researchers and educators to understand the students’ learning processes in an online learning environment. For example, contributions to a discussion forum by an information-giver on a discussion topic can be drastically different from students who seek information. Similarly, significant differences can be observed in the linguistic expressions exhibited by two different learner groups (pass-grade learners and fail-grade learners). The thesis also advances understanding of student learning to an extent by presenting machine learning models, topic models and decision-making rule sets that provide meaningful insights to both students and education providers in an online learning medium, especially MOOCs.