Decentralized Collaborative Filtering Algorithm with Privacy Preserving for Recommendation in Mobile Edge Computing

Abstract

AbstractMobile edge computing (MEC) deploys network services closer to the user’s wireless access network side and provides IT service environment and cloud computing capabilities at the edge of the mobile network. With the advantages of low latency and high bandwidth, many context-aware services (such as recommendation) have been greatly developed. However, standard recommendation architecture in MEC focuses on individual MEC entities and the vertical interaction between end-user and a single MEC node. Its quality of service (QoS) is limited by the performance of a particular edge node, and the resources of edge servers are not fully utilized during the idle period of user activity. What’s more, the user’s behavioral data stored in single edge server will lead to the risk of privacy disclosure. In response to these problems, we propose a decentralized collaborative filtering algorithm in MEC, to amalgamates the heterogeneous resources at the edges and to protect user’s privacy. The core of our structure is composed of user terminals, edge nodes, and cloud nodes. User terminals are the targets of recommendation services, including static and mobile devices, and can undertake a small portion of the recommendation tasks. Edge nodes are the MEC servers co-located with the base stations and are responsible for handling the low latency and computation-intensive tasks. Cloud nodes are traditional mobile cloud computing (MCC) servers located at the remote data center for latency-tolerant and computation-intensive tasks. We provide three observations on the decentralized recommendation in MEC which are from the aspects of latency, resource utilization and user privacy. In our algorithm, the rating data is divided into public and private data and processed with decentralized matrix completion.KeywordsMobile edge computingDecentralized recommendationPrivacy preservingQuality of serviceCollaborative filtering