Date of Award

August 2020

Degree Type


Degree Name

Doctor of Philosophy (PhD)




James Bellini


Big Data, hidden Markov models, interaction patterns, learner interaction, learning analytics, Massive Private Online Courses

Subject Categories



The purpose of this study was to conduct data-driven research by employing learning analytics methodology and Big Data in learning management systems (LMSs), and then to identify and compare learners’ interaction patterns in different achievement groups through different course processes in Massive Private Online Courses (MPOCs).

Learner interaction is the foundation of a successful online learning experience. However, the uncertainties about the temporal and sequential patterns of online interaction and the lack of knowledge about using dynamic interaction traces in LMSs have prevented research on ways to improve interactive qualities and learning effectiveness in online learning. Also, most research focuses on the most popular online learning organization form, Massive Open Online Courses (MOOCs), and little online learning research has been conducted to investigate learners’ interaction behaviors in another important online learning organization form: MPOCs.

To fill these needs, the study pays attention to investigate the frequent and effective interaction patterns in different achievement groups as well as in different course processes, and attaches importance to LMS trace data (log data) in better serving learners and instructors in online learning. Further, the learning analytics methodology and techniques are introduced here into online interaction research.

I assume that learners with different achievements express different interaction characteristics. Therefore, the hypotheses in this study are: 1) the interaction activity patterns of the high-achievement group and the low-achievement group are different; 2) in both groups, interaction activity patterns evolve through different course processes (such as the learning process and the exam process). The final purpose is to find interaction activity patterns that characterize the different achievement groups in specific MPOCs courses.

Some learning analytics approaches, including Hidden Markov models (HMMs) and other related measures, are taken into account to identify frequently occurring interaction activity sequence patterns of High/Low achievement groups in the Learning/Exam processes under MPOCs settings. The results demonstrate that High-achievement learners especially focused on content learning, assignments, and quizzes to consolidate their knowledge construction in both Learning and Exam processes, while Low-achievement learners significantly did not perform the same. Further, High-achievement learners adjusted their learning strategies based on the goals of different course processes; Low-achievement learners were inactive in the learning process and opportunistic in the exam process. In addition, despite achievements or course processes, all learners were most interested in checking their performance statements, but they engaged little in forum discussion and group learning. In sum, the comparative analysis implies that certain interaction patterns may distinguish the High-achievement learners from the Low-achievement ones, and learners change their patterns more or less based on different course processes.

This study provides an attempt to conduct learner interaction research by employing learning analytics techniques. In the short term, the results will give in-depth knowledge of the dynamic interaction patterns of MPOCs learners. In the long term, the results will help learners to gain insight into and evaluate their learning, help instructors identify at-risk learners and adjust instructional strategies, help developers and administrators to build recommendation systems based on objective and comprehensive information, all of which in turn will help to improve the achievements of all learner groups in specific MPOC courses.


Open Access

Included in

Education Commons