Authorization Robot

Beta phase

Role concepts, AI, Evolutionary Algorithms, role mining, semantics

The challenge of implementing the automatic authorization concept is that the new software must be able to process very large amounts of data. These data volumes are too large for the standard procedures for data analysis available in the scientific literature. Efficient matrix clustering methods are used here, which identify related departments/processes/patterns in the usage data as preprocessing, thus allowing a partitioning with as little information loss as possible and reducing the problem size.

Based on this partitioned usage data, an Evolutionary Algorithm is used to generate the best possible role concept for it. This "role mining problem", known from the literature, is assigned to the complexity class NP-complete. Therefore, the evolutionary approach as a metaheuristic method offers the right tool to approach efficiently and steadily an optimal solution. Furthermore, a constant interaction with the algorithm must be guaranteed in order to transfer existing consulting knowledge to the virtual consultant during a learning process.

In order to provide the Evolutionary Algorithm with suitable evaluation criteria for the role concepts, it requires additional complex modules that assist the main algorithm. Compliance can be given as an example here. A set of rules for all audit relevant conflicts and critical authorization combinations that allows the compliance of a role concept to be evaluated and optimized accordingly.

Automated evaluation of the semantic meaningfulness of roles is a much more difficult endeavor. So, are the generated roles also nameable, i.e. could a human give a semantically meaningful name to the role and its content? Here, we use a machine learning method, more precisely neural networks, to learn the experience of consultants and experts from previous projects and other sources, so that this module can evaluate whether a role can be named by a human and, if so, which name makes sense.