基于差分隐私和SVD++的协同过滤算法

协同过滤技术在推荐系统的实现中具有广泛的应用,协同过滤以用户对商品项目的评价分数为依据,而这些评价有可能反映用户某些不欲为人知的喜好特点,因此,对具备隐私保护能力的协同过滤模型的研究引起了普遍的关注.SVD++是当前最为常用的协同过滤模型之一,差分隐私模型则是近十年来隐私保护理论最重要的研究进展之一,将两者相结合提出3种基于差分隐私和SVD++的协同过滤模型:基于梯度扰动的SVD++隐私保护模型、基于目标函数扰动的SVD++隐私保护模型和基于输出结果扰动的隐私保护模型.理论分析和实验结果显示,所提出的算法不仅能为用户的隐私安全提供可靠的保障,而且还可保持较高的预测准确度.

Collaborative filtering via SVD++ with differential privacy

Collaborative filtering(CF), as a technique that automatically predicts the interest of an user by collecting rating information from other similar users or items, has been widely deployed in various recommendation systems. However, CF prediction is based on the users'' historical ratings, indicating that it may reflect some of the users'' private preferences. Consequently, enhancing CF model with privacy preservation guarantee has attracted much research attention. In this paper, we propose three privacy preserving collaborative filtering algorithms: DPSS++, DPSAObj++ and DPSAOut++. All the algorithms are based on SVD++, one of the most used CF algorithms, and differential privacy model, one the most important advance in the area of privacy preserving in the last decade. Our analysis shows that the proposed algorithms not only can provide reliable guarantee in terms of privacy preserving, but also keep high prediction accuracy.