Enhancing Edge Caching Efficiency: Leveraging Social and Context-Aware Popularity Rank Prediction via Transfer Learning

Abstract

Managing network congestion and finite edge cache resources in mobile networks has grown increasingly challenging as content delivery networks (CDNs) have evolved from serving major content providers to supporting a wide variety of user-generated content. Consequently, traditional cellular networks face inefficiencies in achieving optimal cache hit counts using reactive approaches. To address this, proactive caching methods based on content popularity rank prediction emerge as promising solutions for efficiently utilizing cellular cache memory. However, training a machine learning (ML)-based prediction model on extensive cellular traffic, incorporating user social attributes and content features, proves to be computationally intensive and time-consuming. In this research, we propose a transfer learning (TL)-based social and context-aware popularity rank prediction framework to circumvent the need for training the ML model from scratch, thereby reducing computational requirements. The results show that the TL method achieves an acceptable accuracy at the target gNodeBs (gNBs). Moreover, the content-size-informed popularity rank prediction outperforms the state-of -the-art CoPUP method and the classical least recently used (LRU) strategy and achieves a satisfactory cache hit count.