Personalized Online Self-Learning

University of Minnesota Ph.D. dissertation. 2021. Major: Computer Science. Advisors: Jaideep Srivastava, George Karypis. 1 computer file (PDF); 137 pages.With the growth of the internet, online learning platforms such as edX, Coursera, and Udacity have emerged. The Massive Open Online Courses (MOOCs) provided by these learning platforms are changing the landscape of education. The advantage of MOOCs is that they make courses available at a nominal price to students all across the globe. With the ability to reach a large number of learners around the world, MOOCs have made a positive impact on education. In addition, professional learners take these courses to achieve professional and career growth. This increases the audience size of the learning platform. Recent studies have shown that MOOCs have emerged as a disruptive technology with the potential of changing the shape of the existing educational setting. Despite the convenient settings provided by MOOCs, dropout rates on the learning platforms remain elevated. Some learners who drop out report a lack of support by these platforms as a major reason for their disengagement. A factor contributing to this lack of personal guidance is that the online learning platforms follow one-size-fits-all and are not customized for different individuals. Currently, in most of the online education settings student has to determine everything, from what courses to pick to what questions to solve. Instead, an ideal learning system must scaffold the learning process—from initial modeling and coaching-oriented feedback to a gradual release of responsibility to students. Without sustained student input and feedback, their talents, creativity, and efficacy can be overlooked or negated. To tackle this problem, we need to develop systems that support self-learning. Personalized self-learning is defined as a teaching and learning process that assists learners based on the strengths, needs, and interests of individual learners while enhancing the self-learning experience. Massive data generated by online learning platforms have made research in this direction possible. Machine learning and data mining communities are focusing on the application of AI in MOOC education research. The first step leading to the development of personalized systems is to identify the needs of individual learners. In an online education system, we must determine the strengths and weaknesses of learners before customizing the platform to their condition. A system to assess learner's knowledge can also help in proving a justification to learners regarding what they need to focus on or what learning trajectory to follow. Second, we personalize the recommendation of forums to improve the experience of students on MOOCs. The discussion forums have become an open-source venue for sharing the knowledge which generates an auxiliary source of learning. For students taking the online courses, this auxiliary material can help the interested student have a constructive discussion with their peers. However, it is difficult for them to browse through the enormous amount of forums to find the relevant thread of their interest. Lastly, we aid self-learners who join the online learning system for developing a specific skill (such as machine learning) or learning a particular concept. Specifically, we provide them the pre-requisite concepts to master before focussing on their goal concept. We believe that this information can help the learners pick the concepts and videos to watch more intelligently. In addition, we also recommend the next videos for learners to watch based on their interaction behavior in the past. For this, we develop a novel representation learning technique that leverages rich information about their textual content and structural relations between entities. In summary, this thesis contributes towards the development of personalized MOOC platforms, specifically providing the following application 1)Knowledge Assessment to determine the strengths and weaknesses of students, 2) Forum recommendation to recommend relevant forums to students, 3) Concept Pre-requisite Prediction to predict pre-requisite relations between different knowledge concepts, and 4) Learning Path recommendation to recommend the sequence of videos a student needs to pick to achieve their goals