Semi-supervised Clustering for Intelligent User Management

As IT systems become more complex and the number of users increases dramatically, automation of user management tasks becomes a high priority for administrators. Grouping users intelligently based on their interactions with the system can serve two purposes. First, it can supply administrators with a global view of the user pool via meaningful user classes that are based on the actual system usage information. Second, given expressive clusters of users, policies that apply to identified user groups can be developed along with profiles corresponding to particular clusters, alleviating the need to tailor group management tasks to a diverse set of users. Thus, grouping users based on their system usage provides insights about the set of users that can be directly translated into typical user profiles and policies for user groups. We propose to employ state-of-the-art machine learning algorithms to discover user groups automatically based on system usage data. Administrator preferences can be incorporated in the group formation process as feedback that may be expressed as assigning certain users to pre-defined groups, or separating users to belong to different groups. Once the groups are created, classification algorithms will be trained for categorizing users. We expect our approach to result in an adaptive, flexible framework for user management that will lead to optimal resource utilization and will be responsive to the needs of administrators.