Learning path design on knowledge graph by using reinforcement learning

Designing an optimal learning path for specific tasks benefits learners and instructors in education. Many approaches adopt the individual knowledge elements to construct the learning path by question and answer. However, the current methods often fail to consider the relationship between knowledge elements. This paper presents a reinforcement learning (RL) approach to learning path design. More specifically, our method constructs a knowledge graph according to the concept relationship contained in exercises and formulates the RL framework with the predefined graph constraints. To achieve rewards in RL, the proposed formulation employs the deep knowledge tracing model for evaluation. With our formulation, we evaluated the classical RL models, including the Deep Q-network and its variants, to design the learning path. Finally, the simulated experiments are conducted on a public dataset for online education. Experimental results show that the introduced RL approach generates effective learning sequences to improve the correct probability of question answers. This study contributes to pave the path to learning path design with reinforcement learning.