Mindful Interruptions: A Lightweight System for Managing Interruptibility on Wearables

We present the design, development, and evaluation of a personalized, privacy-aware and multi-modal wearable-only system to model interruptibility. Our system runs as a background service of a wearable OS and operates on two key techniques: i) online learning to recognize interruptible situation at a personal scale and ii) runtime inference of opportune moments for an interruption. The former is realised by a set of fast and efficient algorithms to automatically discover and learn interruptible situations as a function of meaningful places, and physical and conversational activities with active user engagement. The latter is substantiated with a multi-phased context sensing mechanics to identify moments which are then utilized to delivery notifications and interactive contents at the right moment. Early experimental evaluation of our system shows a sharp 46% increase in the response rate of notifications in wearable settings at the expense of negligible 6.3% resource cost.