Educational Question Mining At Scale: Prediction, Analysis and Personalization

Online education platforms enable teachers to share a large number of educational resources such as questions form exercises and quizzes for students. With volumes available questions, it is important have an automated way quantify their properties intelligently select them students, enabling effective personalized learning experiences. In this work, we propose framework mining insights from at scale. We utilize the state-of-the-art Bayesian deep method, in particular partial variational auto-encoders (p-VAE), analyze real students' answers collection questions. Based on p-VAE, two novel metrics that question quality difficulty, respectively, strategy adaptively apply our proposed real-world dataset with tens thousands millions online platform. Our not only demonstrates promising results terms statistical but also obtains highly consistent domain experts' evaluation.