Structural-Based Graph Publishing Under Differential Privacy

Mining data from social and communication network have been attracting recent attention from various research fields. However, these data represented by large-scale graphs are often sensitive and private. It is a necessity of developing algorithms to publish large-scale graph while not revealing sensitive information. As a standard for data privacy preservation, differential privacy based algorithm are also widely used in publishing graph-based dataset. However, previous differential privacy based methods often bring huge computational cost and lack the capability of modeling complicated graph structure. To address these challenge, we propose a novel graph publishing algorithm which combines community detection with differential privacy method. By segmenting the graph into several sub-graphs by community detection, differential privacy methods is able to handle large-scale graphs with complex structure. Experimental results on several datasets demonstrates the promising performance of the proposed algorithm compared with original differential privacy methods.