Simulating users to support the design of activity management systems

We describe a simulation system that models the user of a calendar-management tool. The tool is intended to learn the user's scheduling preferences, and we employ the simulator to evaluate learning strategies. The simulated user is instantiated with a set of preferences over local and global features of a schedule such as the level of importance of a particular meeting and the amount of preparation time available before it is to begin. The system then processes a set of simulated meeting requests, and over time and through user feedback, it learns the user's preferences, affording it the ability to thereafter manage the user's schedule more autonomously.