An Accurate and Scalable Role Mining Algorithm based on Graph Embedding and Unsupervised Feature Learning

Role-based access control (RBAC) is one of the most widely authorization models used by organizations. In RBAC, accesses are controlled based on the roles of users within the organization. The flexibility and usability of RBAC have encouraged organizations to migrate from traditional discretionary access control (DAC) models to RBAC. The most challenging step in this migration is role mining, which is the process of extracting meaningful roles from existing access control lists. Although various approaches have been proposed to address this NP-complete role mining problem in the literature, they either suffer from low scalability, or present heuristics that suffer from low accuracy. In this paper, we propose an accurate and scalable approach to the role mining problem. To this aim, we represent user-permission assignments as a bipartite graph where nodes are users and permissions, and edges are user-permission assignments. Next, we introduce an efficient deep learning algorithm based on random walk sampling to learn low-dimensional representations of the graph, such that permissions that are assigned to similar users are closer in this new space. Then, we use k-means and GMM clustering techniques to cluster permission nodes into roles. We show the effectiveness of our proposed approach by testing it on different datasets. Experimental results show that our approach performs accurate role mining, even for large datasets.