An Edge Caching Strategy Based on Separated Learning of User Preference and Content Popularity

Mobile edge caching has recently been proposed as a promising technique to efficiently relieve traffic burden by storing contents in the mobile edge server, or even the base station (BS). However, it remains a challenging problem that how to select the appropriate content considering the limited use...

Full description

Saved in:
Bibliographic Details
Published in2021 IEEE/CIC International Conference on Communications in China (ICCC) pp. 1018 - 1023
Main Authors Chen, Guanpeng, Jing, Wenpeng, Wen, Xiangming, Lu, Zhaoming, Zhao, Shuyue
Format Conference Proceeding
LanguageEnglish
Published IEEE 28.07.2021
Subjects
Base stations
Collaborative filtering
Heuristic algorithms
mobile edge caching
neural network
Performance evaluation
Prediction algorithms
Predictive models
Simulation
user preference prediction
Online AccessGet full text
DOI10.1109/ICCC52777.2021.9580288

Cover

More Information
Summary:Mobile edge caching has recently been proposed as a promising technique to efficiently relieve traffic burden by storing contents in the mobile edge server, or even the base station (BS). However, it remains a challenging problem that how to select the appropriate content considering the limited users and cache capacity of a BS. Motivated by the advance for mining the preference of users in the recommendation system, we propose a proactive edge caching strategy which is based on a novel user preference prediction model called Neural Collaborative Filtering without Content Popularity (N CFCP). Since content popularity and personal preference have different laws of change, our proposed strategy attempts to separate them to improve the accuracy of prediction for user preference. An optimization problem is formulated to maximize the cache hit ratio and is decomposed into two subproblems, i.e., the individual preference prediction and content placement. For individual preference prediction, we design the novel NCFCP method, which can model the relationship between users and contents more precisely. For content placement, we develop a caching mechanism based on the group preference where user activity is considered. Performance evaluation over a real-world dataset shows that the proposed algorithm outperforms baseline algorithms in terms of the cache hit ratio and users' satisfaction by 18% and 20% on average, respectively.
DOI:10.1109/ICCC52777.2021.9580288