Scholarlink

<p>College-level Physical Education (PE) course selection is influenced by diverse student characteristics such as personal interests, physical capabilities, and long-term fitness goals, making personalized recommendations a complex task. Traditional Collaborative Filtering (CF) and Deep Learning (DL)-based recommendation approaches lack the ability to model sequential decision-making, adapt to dynamic behavioral feedback, and optimize long-term learning outcomes, resulting in limited personalization and reduced student engagement. To address these limitations, this study proposes DRL-PRS, a Deep Reinforcement Learning-Based Personalized Recommendation System that incorporates advanced algorithmic components for enhanced decision intelligence. The framework employs a state representation learning module that encodes student profiles, course attributes, engagement histories, and performance trajectories into compact latent vectors, complemented by a reward modeling mechanism that integrates student satisfaction, engagement, and completion likelihood into a composite reward signal. A policy optimization engine based on Actor–Critic methods, Proximal Policy Optimization (PPO) is used to iteratively update the recommendation policy, while an exploration–exploitation controller leveraging ε-greedy and entropy regularization ensures balanced discovery of new course options. Experimental results demonstrate that DRL-PRS significantly outperforms conventional CF (0.65) and DL (0.72) models, achieving 89.2% accuracy, 87.6% precision, 86.1% recall, 86.8% F1-score, and a User Satisfaction Index (USI) of 0.87. Furthermore, the system enhances student engagement by 15.3% and increases course completion rates by 12%, confirming its effectiveness as an adaptive, accurate, and context-aware recommendation solution for PE course selection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep reinforcement learning based personalized recommendation system for college physical education courses

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月16日

查看全文
  • Haibo Cao

摘要

College-level Physical Education (PE) course selection is influenced by diverse student characteristics such as personal interests, physical capabilities, and long-term fitness goals, making personalized recommendations a complex task. Traditional Collaborative Filtering (CF) and Deep Learning (DL)-based recommendation approaches lack the ability to model sequential decision-making, adapt to dynamic behavioral feedback, and optimize long-term learning outcomes, resulting in limited personalization and reduced student engagement. To address these limitations, this study proposes DRL-PRS, a Deep Reinforcement Learning-Based Personalized Recommendation System that incorporates advanced algorithmic components for enhanced decision intelligence. The framework employs a state representation learning module that encodes student profiles, course attributes, engagement histories, and performance trajectories into compact latent vectors, complemented by a reward modeling mechanism that integrates student satisfaction, engagement, and completion likelihood into a composite reward signal. A policy optimization engine based on Actor–Critic methods, Proximal Policy Optimization (PPO) is used to iteratively update the recommendation policy, while an exploration–exploitation controller leveraging ε-greedy and entropy regularization ensures balanced discovery of new course options. Experimental results demonstrate that DRL-PRS significantly outperforms conventional CF (0.65) and DL (0.72) models, achieving 89.2% accuracy, 87.6% precision, 86.1% recall, 86.8% F1-score, and a User Satisfaction Index (USI) of 0.87. Furthermore, the system enhances student engagement by 15.3% and increases course completion rates by 12%, confirming its effectiveness as an adaptive, accurate, and context-aware recommendation solution for PE course selection.