Role mining algorithm evaluation and improvement in large volume android applications

Abstract

Role mining is a very useful engineering method to help administrators set up the mechanism of role based access control for information systems, but not applied in the Android security framework so far. This paper uses large volume Android applications from the Android Market (Google Play Store now), which include 44,971 applications (subjects), 125 permissions, and 222,734 application-permission assignments (application, permission), to evaluate the effectiveness of five popular role mining algorithms: HM, HPr, HPe, GO, and ORCA. Furthermore, according to the features of Android applications, we propose Mine-Tag, an algorithm that generates tags based on the descriptions of Android applications. These tags can be attached to each mined role to help administrators manage the roles. We set up experiments, evaluate algorithms, and discuss the insights of mining methods in Android applications.