Scholarlink

<p>Traditional recommender systems often struggle with challenges such as the cold start problem, data sparsity, and adaptation to changing user preferences. This paper explores the integration of reinforcement learning (RL), specifically multi-armed bandits (MAB), with active learning strategies to address these limitations. MAB-based models dynamically balance exploration and exploitation, allowing continuous learning and enabling the system to adapt to changing user preferences over time. Additionally, active learning enhances personalization by selectively acquiring user feedback, further improving recommendation accuracy. To evaluate the effectiveness of this approach, a comprehensive comparative analysis of various MAB policies and active learning strategies is conducted using the MovieLens ML1M dataset. The results highlight the trade-offs between different strategies, offering valuable insights into various configurations for optimizing recommendation performance. While the integration of MAB and active learning shows significant potential in mitigating data sparsity and cold start issues, the findings highlight the need for further refinement to enhance real-world applicability. This paper contributes to the advancement of reinforcement learning-based recommender systems by demonstrating how continuous learning and active learning mechanisms enhance recommendation efficiency, personalization, and adaptability, highlighting the importance of ongoing innovation in scalable recommendation frameworks.</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

Active learning-based multi-armed bandits for recommendation systems

  • 期刊论文

  • 发表日期：

    2025年6月19日

查看全文
  • Bachir Asri,
  • Sara Qassimi,
  • Said Rakrak

摘要

Traditional recommender systems often struggle with challenges such as the cold start problem, data sparsity, and adaptation to changing user preferences. This paper explores the integration of reinforcement learning (RL), specifically multi-armed bandits (MAB), with active learning strategies to address these limitations. MAB-based models dynamically balance exploration and exploitation, allowing continuous learning and enabling the system to adapt to changing user preferences over time. Additionally, active learning enhances personalization by selectively acquiring user feedback, further improving recommendation accuracy. To evaluate the effectiveness of this approach, a comprehensive comparative analysis of various MAB policies and active learning strategies is conducted using the MovieLens ML1M dataset. The results highlight the trade-offs between different strategies, offering valuable insights into various configurations for optimizing recommendation performance. While the integration of MAB and active learning shows significant potential in mitigating data sparsity and cold start issues, the findings highlight the need for further refinement to enhance real-world applicability. This paper contributes to the advancement of reinforcement learning-based recommender systems by demonstrating how continuous learning and active learning mechanisms enhance recommendation efficiency, personalization, and adaptability, highlighting the importance of ongoing innovation in scalable recommendation frameworks.