Course recommendation of MOOC with big data support: A contextual online learning approach

With the advent of the big data era of MOOC, enrolled students and offered courses become numerous and diverse, resulting in a large amount of data and complex curriculum relationships. Thus how to recommend appropriate course to improve students' learning outcomes has become a daunting task. The state-of-the-art works ignore some significant features in course recommendation of MOOC: heterogeneity of large-scale user groups, sequence problem in courses and foreseeable quantitative explosion of courses and users. This paper proposes a systematic methodology for recommending personalized courses with considering the sequence of learning curriculum. The system works by recommending the course with the highest reward to a user. New feedback of the user is then recorded and will be used to improving the performance of recommendation for future students. The core component is a novel online learning algorithm based on hierarchical bandits with known smoothness. We analyze the performance of our proposed online learning algorithm in terms of regret, and prove the asymptotic optimality of the proposed algorithm. Experimental results are provided to verify our theory.