Rec-DOGA: Recommendation-Enabled Dynamic Online Gradient Ascent Cache Algorithm

Caching integrated with recommendation systems has emerged as a prominent trend in the caching research field, particularly concerning the volume of data within the network and diverse user requirements. We cast the caching problem in dynamic network environments and propose a new approach validated in dynamic network environments that combines recommendation systems with caching strategies: a Recommendation-Enabled Dynamic Online Gradient Ascent Cache Algorithm (Rec-DOGA). We first construct an online gradient ascent cache model with the recommendation mechanism to minimize dynamic regret in the recommendation process. Furthermore, we provide theoretical proof for the stability of this strategy when combining recommendation systems and caching while maintaining sublinear dynamic regret. This demonstrates its potential to achieve optimal performance in dynamic environments. Finally, our approach demonstrates outstanding performance in dynamic environments, particularly when the recommendation accuracy reaches 50%, achieving an approximately 7.05% higher hit ratio compared to the OFTRL algorithm, as demonstrated through experiments involving both synthetic and real-world data.