Toward edge minability for role mining in bipartite networks

Abstract

Abstract Bipartite network models have been extensively used in information security to automatically generate role-based access control (RBAC) from dataset. This process is called role mining . However, not all the topologies of bipartite networks are suitable for role mining; some edges may even reduce the quality of role mining. This causes unnecessary time consumption as role mining is NP-hard. Therefore, to promote the quality of role mining results, the capability that an edge composes roles with other edges, called the minability of edge , needs to be identified. We tackle the problem from an angle of edge importance in complex networks; that is an edge easily covered by roles is considered to be more important. Based on this idea, the k -shell decomposition of complex networks is extended to reveal the different minability of edges. By this way, a bipartite network can be quickly purified by excluding the low-minability edges from role mining, and thus the quality of role mining can be effectively improved. Extensive experiments via the real-world datasets are conducted to confirm the above claims.