Boosting the accuracy of differentially private in weighted social networks

Abstract

Social network not only helps people to build its internet applicable service, but also collects a large amount of user information (i.e., sensitive data), which may reveal potential privacy information by analyzing these data. At present, the differential privacy protection model gives a rigorous, quantitative representation and proof to the risk of privacy disclosure, which greatly ensures the availability of data. MBCI, a stochastic perturbation algorithm based on differential privacy, is designed. First, it uses the undirected weighted graph as the social network, and the sequence of edge weight is treated as an ordered histogram. Then, the buckets with the same count are merged into groups in the histogram and it satisfies the differential privacy by adding the noise to the weights with sensitive information. The shortest path of the network keeps unchanged by consistent reasoning of the original sequence. In order to reduce the more substantial error MBCI generated, we propose a novel algorithm - LMBCI. LMBCI first divides the original weighted social network and then constructs an algorithm under the differential privacy for each sub-network. The experimental results show that LMBCI can effectively reduce the error, improve the accuracy and retain more statistical characteristics compared with MBCI.