基于推荐的社交媒体数据发布的隐私保护

Privacy Protection Based on the Publication of Recommended Social Media Data

个性化的推荐需要使用大量的用户数据,尤其是用户在社交媒体上的活动数据,包括评级、签到等,然而,从大量的用户活动数据中,能够推断出用户的隐私数据。在本文中,针对FM推荐算法的特性,提出距离度量KFC,约束数据失真,提出了PrivFM,一个可定制的、连续的、保护隐私的社交媒体数据发布框架,通过扰乱用户发布的活动数据,防止推理攻击,同时保证推荐效用。实验结果表明,相对于其他的隐私保护方法及距离度量,提高了隐私保护与推荐之间的平衡。

Personalized recommendation requires the use of a large amount of user data, especially the activity data of users on social media, including ratings, checkins, etc. However, from a large amount of user activity data, users’ privacy data can be inferred. In this paper, aiming at the characteristics of FM recommendation algorithm, distance measurement KFC (Kendall feature correlation) was proposed to constrain data distortion, and PrivFM, a customizable, continuous and privacyprotecting social media data publishing framework, was proposed to prevent inference attacks by disrupting the active data published by users, while ensuring the recommendation effectiveness. The experimental results show that compared with other privacy protection methods and distance measurement, the balance between privacy protection and recommendation is improved.