关于分布式网络用户数据隐私协同保护仿真

Simulation of Data Privacy Collaborative Protection for Distributed Network Users

对分布式网络用户数据进行隐私协同保护能够有效提高用户覆盖率、缩短运行时间、提高准确率等。对用户数据隐私协同的保护,先建立网络用户兴趣模型,再利用EM算法最大化似然函数对模型估值,计算用户间相似度,找出目标用户最近邻,完成用户数据隐私协同保护。传统方法将目标函数加扰方法应用于ALS目标函数中,求出这两个真实数据集,但忽略了对用户相似度估算,导致运行时间长。提出一种基于用户模糊相似度的方法。利用梯形模糊数确定网络用户信息量,得到数据梯形模糊评分值,利用该值计算网络用户兴趣数据概率值,通过概率值构建网络用户兴趣模型。引入EM算法最大化似然函数估计模型,用估值计算用户间相似度,以此找出目标用户最近邻,并对所有未评分数据项目进行评分。结果表明,所提方法在分布式数据隐私协同保护中,用户覆盖率高、运行时间短、有较高准确性。

In the traditional method,the objective function scrambling method was applied to ALS objective function,but the estimation of user similarity was ignored,which results in long running time.Therefore,a method based on user fuzzy similarity was proposed.Firstly,the trapezoidal fuzzy number was used to determine the amount of network user information and thus to obtain the trapezoidal fuzzy score.Then,this value was used to calculate the probability value of network user interest data and the network user interest model was constructed through the probability value.In addition,the EM algorithm was introduced to maximize the likelihood function estimation model.Finally, the similarity degree between users was calculated by the estimation value,so as to find out the nearest neighbor of target user.On this basis,all unrated data items were scored.The result proves that the proposed method has high user coverage rate,short running time and high accuracy in distributed data privacy protection."