Privacy-preserving EM algorithm for clustering on social network

Abstract

We consider the clustering problem in a private social network, in which all vertices are independent and private, and each of them knows nothing about vertices other than itself and its neighbors. Many clustering methods for networks have recently been proposed. Some of these works have dealt with a mixed network of assortative and disassortative models. These methods have been based on the fact that the entire structure of the network is observable. However, entities in real social network may be private and thus cannot be observed. We propose a privacy-preserving EM algorithm for clustering on distributed networks that not only deals with the mixture of assortative and disassortative models but also protects the privacy of each vertex in the network. In our solution, each vertex is treated as an independent private party, and the problem becomes an n-party privacy-preserving clustering, where n is the number of vertices in the network. Our algorithm does not reveal any intermediate information through its execution. The total running time is only related to the number of clusters and the maximum degree of the network but this is nearly independent of the total vertex number. © 2012 Springer-Verlag.