A Hybrid Active and Passive Cache Method Based on Deep Learning in Edge Computing

Zhengchang Song, Rui Cao, Bingxin Niu, Junhua Gu, Chunjie Li

Abstract: In recent years, the rapid development of micro-video in the multimedia field has brought about a huge increase in Internet traffic. Ensuring consumer quality of experience (QoE) has become a major challenge for Internet service providers (ISPs). To alleviate the burden of Internet traffic, we propose a hybrid active-passive cache update strategy. For newly released micro-videos, the multimodal transformer popularity prediction model (MTPP) is used to actively predict the popularity of micro-videos. For micro-video files in the base station, a dynamic cache based on the popularity prediction algorithm (DCPP) is used to update the local popularity through changes in local requests. Extensive experiments on public datasets demonstrate that our proposed popularity prediction method outperforms traditional prediction models in the field of micro-video prediction. In the simulation experiment, our proposed dynamic cache algorithm based on popularity prediction outperforms the traditional cache replacement algorithm.