Leveraging lattices to improve role mining

In this paper we provide a new formal framework applicable to role mining algorithms. This framework is based on a rigorous analysis of identifiable patterns in access permission data. In particular, it is possible to derive a lattice of candidate roles from the permission powerset. We formally prove some interesting properties about such lattices. These properties, a contribution on their own, can be applied practically to optimize role mining algorithms. Data redundancies associated with co-occurrences of permissions among users can be easily identified and eliminated, allowing for increased output quality and reduced processing time. To prove the effectiveness of our proposal, we have applied our results to two existing role mining algorithms: Apriori and RBAM. Application of these modified algorithms to a realistic data set consistently reduced running time and, in some cases, also greatly improved output quality; all of which confirmed our analytical findings. © 2008 Springer Science+Business Media, LLC.