DP2G_(sister):差分隐私社交网络图发布模型

DP2G_(sister): differential privacy publishing model for social network graph

基于社交网络的用户建模、信息推荐等研究近年来得到快速发展,但社交网络中包含大量的个人敏感信息,直接发布此类数据会严重危害个人隐私,因此,在数据发布前进行有效的隐私保护处理具有重要的研究价值和现实意义。基于马尔科夫链蒙特卡洛(Markov Chain Monte Carlo,MCMC)算法,提出了保护数据免受关系推断攻击的隐私保护数据发布方法 DP2G_(sister)和multiR_DP2G_(sister)。DP2G_(sister)首先利用满足差分隐私约束的MCMC算法对原始社交网络图进行采样,进而计算采样结果的连接概率,最终生成重构的具有隐私保证的社交网络发布图。扩展该方法得到的multiR_DP2G_(sister)适用于增量数据发布中的隐私保护。在两个真实的社交网络数据集上进行的实验结果表明,提出的方法能够兼顾数据可用性与隐私性,且在增量数据发布的隐私保护中也能显示出良好的效果。

The research on social networks based user profiling and information recommendation obtained rapid development in recent years. However, social networks contain a large amount of sensitive personnel information, directly releasing such kind of sensitive information which affects the individual privacy seriously. Therefore, studying on privacy preserving before releasing the social network data has significant impact from viewpoints of both research and practical application. In this paper, based on MCMC （Markov Chain Monte Carlo） algorithm, we propose privacy preserving data publication methods, named DP2Gsister, and muhiR DP2Gsisterwhich protect privacy data from relationship inference at- tack. DP2Gsister, algorithm firstly applies MCMC algorithm which satisfies differential privacy constraint to sample the original social network graph, and then calculates the connection probability of the sampling results. Finally, the algorithm generates the reconstructed social network graph with privacy guarantee before data release.