we build a student simulator with our concept-aware deep knowledge tracing model, and then use it to train a flexible and scalable personalized exercise recommendation policy with deep reinforcement learning
We propose a new exercise-level deep knowledge tracing model whose structure is built based on the course's concept list, and the exercise-concept mapping relationships are utilized during students' knowledge tracing. The model supports more input features and obtains higher performance compared with existing models.
We propose an exercises recommendation algorithm which uses model-free reinforcement learning with neural network function approximation to learn an exercise recommendation policy. The policy directly operates on raw observations of a student's exercise history. Experimental results show that our policy achieves better performance than existing heuristic policy in terms of maximizing students' knowledge level.