Get A Sense of Accomplishment in Doing Exercises: A Reinforcement Learning Perspective
Songdeng Niu, Sheng Cao · 2022 IEEE 25th International Conference on Computer Supported Cooperative Work in Design (CSCWD) · 2022
Doing exercises is vital to improving students’ learning effects. Setting reasonable rewards in the process of doing exercises can promote students’ problem-solving interests. Reinforcement learning has been employed on optimal instructional sequencing in education for decades. Existing reinforcement-learning-induced policies in instructional sequencing mainly pay attention to how students take actions in learning activities while attention to the setting of rewards in doing exercises has been little. To redress this imbalance, we first propose a set of calculation formulas of reward function closely associated with the difficulty of each exercise item and the consuming-time of each student. We put forward the confidence index of students as another important parameter in our reward function, which represents how sure each student is that he/she could do the exercises right. We then propose and prove a reward shaping scheme based on dynamic potential function, which can not only ensure invariance of the optimal policy of acquiring exercise-doing rewards but also accelerate the convergence of the exercise-doing process. This scheme provides a new paradigm for motivating students to do exercises in the e-learning environment, which helps to enhance students’ learning enthusiasm, so as to improve their learning efficiencies impressively.