一种基于差分隐私的个性化服务推荐算法

Personalized service recommendation algorithm based on differential privacy

推荐系统存在用户隐私安全性低、推荐服务质量差的问题。为此,文中提出一种基于差分隐私的个性化服务推荐算法DPk⁃median。该算法针对推荐系统服务器中生成的推荐列表,首先利用k⁃median聚类算法将推荐数据中具有相同属性的数据进行聚类;然后根据不同簇的风险级别,添加相应的拉普拉斯噪声机制,同一簇中的隐私预算参数是相同的,在保证隐私的前提下,可合理控制噪声的加入并提高噪声的利用率,保证推荐的质量损失减小,同时增加算法的执行效率。相关实验结果表明,与以往的基于差分隐私的个性化服务推荐系统相比,文中所提出的算法在保证系统安全性的同时,提高了服务推荐的质量和算法的执行效率。

As the recommendation system has the problems of the low user privacy security and poor recommendation service quality,a personalized service recommendation algorithm DPk⁃median based on differential privacy is proposed.In the algorithm,for the recommendation list generated in the recommendation system server,the k⁃median clustering algorithm is used to cluster the data with the same attributes in the recommendation data.The corresponding Laplace noise mechanism is added into cluster with different risk level,among which the privacy budget parameters in the same cluster are the same.On the premise of ensuring privacy,the addition of noise can be reasonably controlled,and the utilization of noise can be improved,so as to reduce the recommended quality loss and increase the execution efficiency of the algorithm.The relevant experiments show that in comparison with the previous personalized service recommendation system based on differential privacy,the proposed algorithm not only ensures the system security,but also improves the quality of service recommendation and the execution efficiency of the algorithm.