Congratulations, Fareedah!
Dr. Fareedah Alsaad has passed her final Ph.D. exam. Her dissertation addresses an important problem about the current learning methods offered by Massive Open Online Courses (MOOCs) platforms. Dr. Alsaad's used data mining and machine learning methods to build the initial infrastructure needed to offer customized learning in MOOCs platform. Here's the abstract of her dissertation.
Massive Open Online Courses (MOOCs) have gained popularity worldwide, with the numbers of courses, institutions, and learners proliferating each year. While MOOCs have provided learners with easy access to numerous well-developed courses, their service resembles the traditional university in terms of the curricula provided or the way success is measured. MOOC platforms provide learners with degree programs and courses from various institutions that reward learners with certificates determining their success. This type of formal education is useful for learners with the intended goal of improving their knowledge and earning certificates as proof of credentials. However, MOOC learners have a variety of learning goals when joining courses other than just completing the course content and receiving certificates. According to the literature, learners registered for MOOCs for many reasons: exploring the course, learning specific concepts within a MOOC, supplementing current academic endeavors, etc. When learners achieve their needs and goals, they leave the course and hence they are considered dropouts according to the traditional definition of success.
Despite the diversity of learners' goals, MOOC platforms offer only certificate-driven style service. The design of current MOOC platforms lacks support to learners who want to utilize MOOCs as modularized resources instead of regular courses. Learners need to manually locate the educational materials that address their learning goals which can be time-consuming. To support the diverse needs of MOOC learners, we need to adopt a model that shifts the online learning landscape from delivering only course-centered learning to providing goal-aware personalized learning paths that satisfy learners' diverse needs. Yet, before we can offer customized learning plans, we must first model the content structure of MOOCs so that we capture the topics offered by different courses and how these topics relate to each other. In this thesis, we introduce two MOOCs' knowledge structure representations: the Concept Dependency Graphs and the Topic Transition Maps and discuss the approaches we developed to construct these graphs. Our main technical innovation lies in exploiting the temporal ordering amongst lectures and concepts to discover the graphs. Since experts manually design the chronological order of lectures, lecture sequences encapsulate rich information about how concepts and topics are related. Therefore, lecture sequences would help in modeling beneficial relations such as dependency and precedence relations. By leveraging the time ordering among and within lectures and also globally throughout the course delivery, we developed two measures: the Bridge Ensemble Measures and the Global Direction Measure to construct the concept dependency graph. The concept dependency graph models the prerequisite relations between concepts, and hence it facilitates finding the prerequisite concepts required to learn any desired concept. In contrast, the topic transition map is a dynamic structure that models the precedence relations between topics and thus captures the variations of different learning paths followed by various courses. As a result, the topic transition map can be the initial block for many applications that recommend customized learning paths to learners. Our approach to construct the topic transition map incorporates clustering methods with the temporal features of lectures to cluster similar lectures into topics and learn the transitions between topics. By relying on the two knowledge structures, we introduce a customized learning approach; CustomLearn, to facilitate goal-oriented learning in MOOC platforms. Our approach of customizing learning proposes four learning goals that can be useful for learners: Exploring (explore a subject), Conforming (learn popular topics in a subject), Mastering (be expert in a subject), and Personalizing (create personalized study plan). These learning goals use subjects to deliver customized learning plans that span multiple courses. Thus, the customized learning approach organizes similar courses in such a way that makes the educational materials easily accessible by learners. To assess our proposed customized learning approach, we developed an early prototype of CustomLearn. We utilized the prototype to conduct a user study with MOOC learners to evaluate the viability of CustomLearn. Our preliminary evaluation indicated that CustomLearn is a viable approach; evident by participants' high perceptions of the usefulness of CustomLearn and their willingness to use and recommend it if it is offered by MOOC platforms. We also evaluated learners' acceptability of heterogeneous study plans (i.e., learning curricula that includes lectures offered by different courses). Findings from our user study suggest that most participants found heterogeneous study plans acceptable for learning any topic as long as these plans maintain high instructional quality. Facilitating customized learning in MOOC platforms by integrating goal-oriented learning is a new research problem and our work in this thesis opened up many promising directions. More importantly, identifying the diverse goals and needs of learners and how to accommodating them is a crucial part to customizing learning. Additionally, developing approaches that automatically map learners' demands to study plans is an open problem that need to be further explored.