一种面向图数据的AWG-LDP局部差分隐私保护算法研究

针对传统的图数据隐私保护方法只关注保护属性或结构两者之一易导致节点或边隐私信息泄露的问题，提出了一种对属性加权图的局部差分隐私的保护算法（AWG-LDP）。首先，该算法利用GN算法将图数据划分成社区子图；其次，分别计算每个社区子图的局部敏感度，对于划分后的每一个子图，通过结合结构相似性和属性相似性并添加拉普拉斯噪声进行边扰动，实现局部差分隐私；最后，利用属性泛化的方式将待发布的节点进行泛化，防止节点敏感信息被攻击。利用真实的图数据集进行了不同参数配置以及不同算法的对比实验，实验结果表明该算法提升了隐私保护效果，同时，降低了信息损失，提高了数据的可用性。

Study on AWG-LDP local differential privacy protection algorithm for graph data

Aiming at the problem that traditional graph data privacy protection methods only focus on protecting one of the attributes or structures, which can easily lead to the leakage of node or edge privacy information, this paper proposed a local differential privacy protection algorithm for attribute-weighted graphs(AWG-LDP). Firstly, the algorithm used the GN algorithm to divide the graph data into community subgraphs. Secondly, it calculated the local sensitivity of each community respectively. For each subgraph after division, it realized local differential privacy by combining structural similarity and attribute similarity and adding Laplace noise for edge perturbation. Finally, it generalized the nodes to be published by means of attribute generalization to prevent node sensitive information from being attacked. Using the real graph data set, the paper configured different parameters and compared algorithms with others. The experimental results show that the algorithm improves the privacy protection effect. Meanwhile, it reduces the information loss and improves the availability of data.