一种移动边缘环境下面向隐私保护QoS预测方法

Privacy Protection QoS Forecasting in Mobile Edge Environment

为在移动边缘环境下快速并准确预测Web Service的服务质量,并有效保护用户隐私,提出了一种移动边缘环境下面向隐私保护的QoS预测方法(Edge-Laplace QoS).充分考虑移动边缘环境下用户移动性强、QoS属性值实时性强等问题.首先,利用地理位置信息定位边缘服务器并采用子矩阵划分法得到边缘服务器-QoS数据综合数据集.接着,为保护用户数据的私密性,采用不断更新噪声值改进的差分隐私法完成对边缘环境下原始数据的动态伪装.最后,在预测过程中,以所访问的边缘服务器为中心,利用地理位置信息不断扩大寻找相似用户的范围,从而在移动边缘环境下基于协同过滤法预测用户的属性值.网络开源数据的实验表明,本方法优于传统环境下的预测效率且相比传统隐私保护法更适合在移动边缘环境下进行QoS预测.

In order to fast and accurately forecast Quality of Service(QoS)of different Web Services in Mobile Edge Environment,and effectively protect user privacy,we propose Edge-Laplace QoS(QoS forecasting with Laplace Noise in Mobile Edge Environment),a novel privacy protection oriented QoS forecasting approach in mobile edge environment.Considering the user has strong mobility and real-time of QoS attribute values in mobile edge environment.First,edge servers are gotten by geographic location information and the dataset of edge server-QoS data is obtained by submatrix partition.Region division is implemented upon latitude and longitude values.It is mainly divided into three steps.Step 1:we target the latitude and longitude values of the geographic locations in the data set.Step 2:the non-repeated positions are selected as the locations of the edge servers.Step 3:the edge region is formed according to the density distribution of the edge servers.We do not need to consider the impact of environmental factors on data set for QoS prediction in traditional environments.However,the existing QoS data set is inapplicable for this research,as the edge environment has the characteristics of fast response and real-time dynamics.We use the following method to integrate the QoS attribute values with the edge servers.First,the data of each QoS attribute is organized in the form of a two-dimensional matrix,where the rows correspond to services and the columns correspond to service users.Each column refers to the QoS attribute values of the service set invoked by a single user,which is treated as a column vector.Second,this matrix is divided into s sub-matrices which might be single-column or multicolumn matrices according to users’actual service accessing records in each edge server,where s is the number of edge servers.This sub-matrix setting can be used to locate the exact QoS value of a specific service that is provisioned over a specific edge server and accessed by a specific user.Next,we also protect the privacy of user data,differential privacy(DP)is improved in the edge environment.Our improved differential privacy method takes the QoS data set in the mobile edge environment as input and continuously updates the random numbers generated by the column vectors in the data set.The dynamic random number is able to enhance privacy protection by adapting to the mobile edge environment.In the training process,we continuously adjust the privacy parameter to achieve its optimal value when the prediction error is minimum.When the minimum prediction error is achieved,the disguised data is closest to the real value.In this paper,we use dynamic Laplace noise to further improve the integrated edge service-QoS data set,and achieve the goal of accurate and efficient QoS forecasting in the mobile edge environment.Third,the location of visited edge server is used as the center,and the geographic location information is used to continuously expand the range of similar users,so as to predict the user’s attribute value using the collaborative filtering method.Experiments are conducted based on several public network data sets.The experimental results demonstrate that Edge-Laplace QoS is superior to the traditional prediction efficiency and is more suitable for edge environment forecasting than the traditional privacy protection method.