一种基于差分隐私保护的协同过滤推荐方法

发布时间：2018-03-18 23:15

本文选题：差分隐私　切入点：隐私保护　出处：《计算机研究与发展》2017年07期 　论文类型：期刊论文

【摘要】：由于推荐系统需要利用大量用户数据进行协同过滤,会给用户的个人隐私带来相当大的风险,如何保护隐私数据成为推荐系统当前面临的重大挑战.差分隐私作为一种新出现的隐私保护框架,能够防止攻击者拥有任意背景知识下的攻击并提供有力的保护.针对推荐系统中的隐私保护问题,提出一种满足差分隐私保护的协同过滤推荐算法.首先,构建用户和项目的潜在特征矩阵,有效降低数据稀疏性;然后,采用目标扰动方法对矩阵中添加满足差分隐私约束的噪声得到噪矩阵分解模型;通过随机梯度下降算法最小化相关联的正则化平方误差函数来获取模型中的参数;最后,应用差分隐私矩阵分解模型进行评分预测,并在MovieLens和Netflix数据集上对算法的有效性进行评价.实验结果证明:所提出方法的有效性能够在有限的精度损失范围内进行推荐并保护用户隐私.
[Abstract]:Because the recommendation system needs to use a large amount of user data for collaborative filtering, it will bring a considerable risk to the privacy of the user. How to protect privacy data has become a major challenge for recommendation systems. Differential privacy is a new privacy protection framework. It can prevent attackers from attacking with arbitrary background knowledge and provide powerful protection. Aiming at privacy protection in recommendation systems, a collaborative filtering recommendation algorithm is proposed to satisfy differential privacy protection. The latent feature matrices of users and projects are constructed to reduce the data sparsity effectively, and then the noise matrix decomposition model is obtained by adding noise satisfying the difference privacy constraints to the matrix by using the objective perturbation method. Random gradient descent algorithm is used to minimize the associated regularization square error function to obtain the parameters of the model. The effectiveness of the proposed method is evaluated on the MovieLens and Netflix datasets. The experimental results show that the proposed method can be recommended within a limited range of loss of accuracy and protect the privacy of users.
【作者单位】：北京工业大学计算机学院;
【基金】：国家自然科学基金项目(91646201,91546111,60803086) 国家科技支撑计划项目(2013BAH21B02) 北京市自然科学基金项目(4153058,4113076) 北京市教育委员会科技计划重点项目(KZ20160005009);北京市教育委员会科技计划一般项目(KM201710005023)~~
【分类号】：TP391.3

【参考文献】