IRMAOC: an interpretable role mining algorithm based on overlapping clustering

Abstract The Industrial Internet motivates the research and development of Zero-Trust Architecture (ZTA). Role-Based Access Control (RBAC) as one of the key technologies for ZTA has become a hot topic. Role mining algorithms are crucial for RBAC and interpretable role mining receives wide attention due to its virtue of mining meaningful roles. However, the roles generated by existing algorithms have low interpretability and high time complexity, limiting their application in practice. This paper proposes an Interpretable Role Mining Algorithm Based on Overlapping Clustering (IRMAOC). It evaluates the interpretability of a role based on user similarity calculated on the permission and attribute of the role, and employs policy interpretability as the metric of a role set. IRMAOC creates a user association graph and clusters to generate candidate roles based on the graph. Then it remains the roles whose interpretability is higher than the preset threshold and re-clusters the users belonging to the other roles until the interpretability of all roles are higher than the threshold. Experimental results show that our algorithm significantly improves the interpretability of roles, reduces the Weighted Structure Complexity (WSC), and decreases time complexity compared to previous works.