Content Caching-Oriented Popularity Forecast Algorithm Design
Qi Chen, Wenze Gao, Takayuki Nakachi, Yitu Wang, Juinjei Liou · 2025
To enable proactive content caching, file popularity forecast becomes an indispensable technique. Conventionally, the objective of information forecast lies in maximizing the accuracy, while neglecting other important metrics, such as forecast confidence and model complexity, which are crucial for the design of content caching. In this paper, we tailor Gaussian Process (GP)-based forecast algorithm for content caching so as to further improve the caching performance. Specifically, we analytically derive the influence of forecast confidence on caching performance, and propose the idea of Controlled Linear Model of Coregionalization (CLMC) to achieve a desired trade-off between forecast confidence and model complexity in terms of minimizing cache fetching loss. The performance improvement is verified by simulation.