k-Anonymity in Federated Heterogenous Graphs and k-Core Anonymization

k-Anonymity is widely used to preserve privacy of individuals by ensuring at least k data points are indistinguishable from one-another. In this paper, we apply k-anonymity to our federated heterogeneous graph storage technique, category cluster. This application of k-anonymity allows select graph edges that would otherwise have to be encrypted with homomorphic encryption to remain in cleartext. We also demonstrate k-core anonymization, a graph anonymization technique based on k-core decomposition. We benchmark anonymized versions of Youtube and LiveJournal social networks against their original counterparts. This anonymization method preserves core membership well (less than 0.7% induced error). k-core anonymization also preserves the distribution of longest paths; however, our results show this distribution is shifted in phase from the original.