Combined fuzzy clustering and firefly algorithm for privacy preserving in social networks

摘要

In recent years, an explosive growth of social networks has been made publicly available for understanding the behavior of users and data mining purposes. The main challenge in sharing the social network databases is protecting public released data from individual identification. The most common privacy preserving technique is anonymizing data by removing or changing some information, while the anonymized data should retain as much information as possible of the original data. K-anonymity and its extensions (e.g., L-diversity and T-closeness) have widely been used for data anonymization. The main drawback of the existing anonymity techniques is the lack of protection against attribute/link disclosure and similarity attacks. Moreover, they suffer from high amount of information loss in the released database. In order to overcome these drawbacks, this paper proposes a combined anonymizing algorithm based On K-member Fuzzy Clustering and Firefly Algorithm (KFCFA) to protect the anonymized database against identity disclosure, attribute disclosure, link disclosure, and similarity attacks, and significantly minimize the information loss. In KFCFA, at first, a modified K-member version of fuzzy c-means is utilized to create balanced clusters with at least K members in each cluster. Then, firefly algorithm is performed for further optimizing the primary clusters and anonymizing the network graph and data. To achieve this purpose, a constrained multi-objective function is introduced to simultaneously minimize the clustering error rate and the generated information loss, while satisfying the defined anonymity constraints. The proposed methodology can be utilized for both network graph structures and micro data. Simulation results over four social network databases from Facebook, Google+, Twitter and YouTube demonstrate the efficiency of the proposed KFCFA algorithm to minimize the information loss of the published data and graph, while satisfying K-anonymity, L-diversity and T-closeness conditions. (C) 2019 Elsevier Ltd. All rights reserved.