Research On Role Mining Algorithm Using Boolean Matrix Factorization

With the explosive growth of the network users,thousands of user data appear on the Internet,and data security issues have received more and more attention.Among many access control technologies,Role-Based Access Control(RBAC),which is the most common access control model,realizes the separation of users and permissions by setting up roles,and reduces the costs of access permission management.How to construct roles is the core issue in RBAC.Role mining,as a main way to automatically construct roles using access control information,is one of the research focuses in RBAC models currently.However,the primary goal of most role mining algorithms is to reduce the complexity of the RBAC policy,while ignoring the real meaning of the role.These mining algorithms will not be adopted because security administrator is unable to understand the role real meaning.Due to the inherent defects of some algorithms which consider the role interpretability problem,there are often problems of low efficiency and low interpretability in large-scale systems.To solve the shortcomings of existing role mining algorithms,we will start with the definitions of the role mining problem(RMP)and the role interpretability,analyze the requirements in different scenarios of RMP,and then model the problem of mining interpretable roles.Two role mining algorithms based on Boolean Matrix Factorization is proposed.The main work is as follows:(1)Basic Interpretable Role Mining Algorithm is proposed.In view of the characteristic of the Basic Role Mining Problem that satisfies the precise matching of permissions,we give optimization goals and constraints combined with the related reserch on role interpretability,and then propose an algorithm which is based on Boolean Matrix Factorization to mine interpretable role in the context of Basic Role Mining Problem.The effectiveness of the algorithm is proved on the simulated data.And compared with other algorithms on the real data,the algorithm proposed improves role interpretability,and has advantages on the number of roles and the running time.(2)Minimal Noise Interpretable Role Mining Algorithm is proposed.In view of the characteristic of the Minimal Noise Role Mining Problem that satisfies the approximate matching of permissions,an algorithm which is aiming to optimizing reconstruction error and role interpretability based on Boolean Matrix Factorization is proposed in the context of Minimal Noise Role Mining Problem.The effectiveness of the algorithm on the simulated data is proved.And compared with other algorithms on the real data,the algorithm we propose improves the role interpretability,effectively reduces the reconstruction error,and has obvious advantages in running time on large data sets.(3)The initial role generation algorithm which generates initial roles for two role mining algorithms is proposed based on the similarity of user attributes and permissions.This algorithm reduces the running time of two role mining algorithms and improves the role interpretability.