K-anonymity for social networks containing rich structural and textual information

Abstract

When social networks are released for analysis, individuals’ sensitive information (e.g., node identities) in the network may be exposed. To avoid unwanted information exposure, social networks need to be anonymized before they are published. In the literature, many approaches exist to anonymize social networks to prevent attacks by adversaries that know the network structures such as node degrees and neighbors. However, these techniques cannot prevent the leakage of valuable identification information during social network analysis if the social network graphs contain both structural and textual information. In this paper, we study the problem of anonymizing social networks to prevent individual identifications which use both structural (node degrees) and textual (edge labels) information in graphs. We formally define the problem as Structure and Text aware \(K\)-anonymity of social networks (STK-Anonymity). In an STK-anonymized network, each individual is \(ST\)-equivalent to at least \(K-1\) other nodes. The major challenge in achieving STK-Anonymity comes from the correlation of edge labels, which causes the propagation of edge anonymization. It has been shown that it is intractable to optimally \(K\)-anonymizing the label sequences of edge-labeled graphs. To address the challenge, we present a two-phase approach which consists of two heuristics in the first phase to process partial graph structures (node degrees in particular) and a set-enumeration tree-based approach in the second phase to anonymize edge labels. Results from extensive experiments on both real and synthetic datasets are presented to show the effectiveness and efficiency of our approaches.