Salt Lake City, Utah
June 23, 2018
June 23, 2018
July 27, 2018
Computers in Education
MOOCs (Massive Open Online Courses) attract a diverse and large set of learners, with largely unknown learning needs and expectations. Researchers have begun to explore reasons learners enroll in MOOCs and have found that learners enroll for variety of reasons and have differing levels of prerequisite knowledge. A new approach to understanding the complex grouping of learners is needed. Commonly used clustering algorithms for large datasets do not take into account all of the dimensions a given learner has. In this paper, we present a computationally fast method to identify significant categories of learners based on their responses to pre-course surveys. The purpose of this paper is to test a recently developed modeling and clustering technique, 1-Dimensional Random Projection (RP1D) with survey data from MOOCs. The RP1D method was developed for high-dimensional data and has been used to find patterns in machine generated learner data. However, the approach has not been tested with survey data. The modeling technique proposed allows one to combine different survey questions and deal with missing responses in a robust fashion through use of a rubric. In order to understand the feasibility and appropriateness of this approach to creating learner groups, we ask, “To what degree does the modeling and RP1D clustering technique result in interpretable groups of MOOC learners?” We apply the technique on pre-course survey data acquired in four MOOCs of the following topics, thermodynamics, math puzzles, forensic science, and mindfulness. The survey asked learners questions concerning their goals for the course and their intended participation in Likert-style statements. Applying the modeling and RP1D method resulted in distinguishable groups of learners in each course according to multiple randomly generated criteria. Depending on the course, 2-6% of the random criteria successfully separated the learners into two distinct groups. For example, in the forensic science course, we found 31 criteria (of 1000) that identified significant groupings of learners. Among the best separated of these learner groups was a 57%/43% split based mainly on differences in the learners’ extrinsic motivation dimensions. A different criterion from that same course found a grouping based mainly on their lifestyle dimensions (a 70%/30% split). Additionally, these criteria persist between the STEM and non‑STEM courses. That is, we found learners grouped into similar clusters regardless of course topic. The ability to separate learner types into distinct categories within and across courses is an important step in furthering the goal of enabling MOOC designers to better design online open educational systems to serve their diverse set of learners. The RP1D technique and implications will be further discussed in the paper.
Williams, T. V., & Douglas, K. A., & Yellamraju, T., & Boutin, M. (2018, June), Characterizing MOOC Learners from Survey Data Using Modeling and n-TARP Clustering Paper presented at 2018 ASEE Annual Conference & Exposition , Salt Lake City, Utah. https://peer.asee.org/30186
ASEE holds the copyright on this document. It may be read by the public free of charge. Authors may archive their work on personal websites or in institutional repositories with the following citation: © 2018 American Society for Engineering Education. Other scholars may excerpt or quote from these materials with the same citation. When excerpting or quoting from Conference Proceedings, authors should, in addition to noting the ASEE copyright, list all the original authors and their institutions and name the host city of the conference. - Last updated April 1, 2015