Scholarlink

Given the shortcomings of existing edge caching techniques in handling personalized user preferences, dynamic content repository updates, and social relationship integration, especially in the context of increasing quality of service (QoS) requirements in sixth-generation (6G) network environments, this study proposes a new approach called Social-Aware Spatio-Temporal Predictor Caching Decision (SASTP-CD). The study aims to enhance user experience and QoS by optimizing caching decisions through more accurate content prediction. The novel proposed method combines a Fourier Graph Neural Network (FourierGNN) with a Kolmogorov-Arnold Network (KAN) to capture spatial-temporal characteristics and social impacts of user requests via multi-kernel residual graphical attention network (MKR-GAT); and proposes a multi-agent proximal policy optimization (Soft MAPPO-TS) algorithm based on soft selection and Thompson sampling for efficient caching decisions. Experimental results show that the proposed SASTP method performs better in predicting user preferences compared to the benchmark method, and SASTP-CD significantly outperforms other evaluated caching strategies in terms of cache hit rate and transmission delay, providing strong support for future Internet of Things (IoT) applications.

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

Optimizing Collaborative Edge Caching in IoT via Social-Aware Spatio-Temporal Prediction and Multi-agent Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ze Liu,
  • Wei Zhang,
  • Hao Hao,
  • Huiling Shi

摘要

Given the shortcomings of existing edge caching techniques in handling personalized user preferences, dynamic content repository updates, and social relationship integration, especially in the context of increasing quality of service (QoS) requirements in sixth-generation (6G) network environments, this study proposes a new approach called Social-Aware Spatio-Temporal Predictor Caching Decision (SASTP-CD). The study aims to enhance user experience and QoS by optimizing caching decisions through more accurate content prediction. The novel proposed method combines a Fourier Graph Neural Network (FourierGNN) with a Kolmogorov-Arnold Network (KAN) to capture spatial-temporal characteristics and social impacts of user requests via multi-kernel residual graphical attention network (MKR-GAT); and proposes a multi-agent proximal policy optimization (Soft MAPPO-TS) algorithm based on soft selection and Thompson sampling for efficient caching decisions. Experimental results show that the proposed SASTP method performs better in predicting user preferences compared to the benchmark method, and SASTP-CD significantly outperforms other evaluated caching strategies in terms of cache hit rate and transmission delay, providing strong support for future Internet of Things (IoT) applications.