计算机应用研究2025，Vol.42Issue(11)：3257-3264,8.DOI:10.19734/j.issn.1001-3695.2025.04.0129
通信高效的个性化联邦多臂赌博机推荐框架
Communication-efficient personalized federated multi-armed bandit recommendation framework
摘要
Abstract
The present work addresses the challenges of data heterogeneity,privacy preservation,fast communication and scala-bility faced in existing content-based personalized recommendation systems.This paper proposed a federated learning algorithm called FTMAB.The algorithm employed a federated learning framework to ensure privacy preservation and manages data hetero-geneity by performing global aggregation of local models through multi-armed bandit techniques.The architecture utilized an upper confidence bound method on the server side for global arm screening recommendation,and optimized communication through a dynamic client-side sampling strategy to aggregate user utility scores on the local client side to enhance the personali-zation of the recommendation.Theoretical analysis proves that the upper bound of regret for FTMAB is O(log T).Experiments on both synthetic and real datasets demonstrate that FTMAB consistently exhibits low regret values while concurrently achieving substantial reductions in communication cost and running time in comparison with existing methodologies.FTMAB framework adeptly balances privacy protection,recommendation quality and communication efficiency,thereby providing an effective solu-tion to the challenges posed by data heterogeneity and scalability in the context of personalized recommendation systems.关键词
个性化推荐/数据隐私/多臂赌博机/数据异质性/联邦学习Key words
personalized recommendation/data privacy/multi-armed bandit/heterogeneous data/federated learning分类
计算机与自动化引用本文复制引用
陈家晟,秦航..通信高效的个性化联邦多臂赌博机推荐框架[J].计算机应用研究,2025,42(11):3257-3264,8.基金项目
湖北高校2020省级科研项目(2020418) （2020418）
湖北省自然科学基金资助项目(2024AFB851) （2024AFB851）
