BusyBody: Creating and Fielding Personalized Models of the Cost of Interruption

Interest has been growing in opportunities to build and deploy
statistical models that can infer a computer user’s current
interruptability from computer activity and relevant contextual
information. We describe a system that intermittently asks users to
assess their perceived interruptability during a training phase and
that builds decision-theoretic models with the ability to predict the
cost of interrupting the user. The models are used at run-time to
compute the expected cost of interruptions, providing a mediator
for incoming notifications, based on a consideration of a user’s
current and recent history of computer activity, meeting status,
location, time of day, and whether a conversation is detected.