Enhancing Math Problem Recommendations Using Reinforcement Learning

• In Event: Virtual Poster Session
  In Virtual iPresentation Only: SIG-Advanced Technologies for Learning Virtual Poster Session

Abstract

This paper explores the application of Deep Q-Networks (DQN) for recommending math problems on the Math Nation platform. By utilizing student interaction data, we developed a DQN-based system that achieved an accuracy of 80% in recommending suitable problems. The study also compared the DQN approach with traditional collaborative filtering and content-based filtering systems, which achieved accuracies of 75% and 17%, respectively. The findings suggest that reinforcement learning offers a more dynamic and personalized recommendation system, effectively addressing the complex dependencies in educational content and learning patterns.

Authors

• Wanli Xing, University of Florida

• Fan Zhang, University of Florida