Reducing distraction of smartwatch users with deep learning

Jemin Lee, Jinse Kwon, Hyungshin Kim

Abstract: Smartwatches are overloaded with various notifications from smartphones. Users are largely distracted, while they may benefit from these relayed notification. To reduce smartwatch user's distraction, we propose an intelligent notification delivery system that relays only important notifications to the smartwatch. We claim that important notifications should be handled within a certain time and they are involved in launching mobile applications. To build model, we collect 6491 notifications and sensor data from three users. A mobile application has been developed to unobtrusively monitor relevant data Then, we implemented a binary classifier which identifies important notifications using deep learning and 8 features are extracted from sensor data. Our classifier shows that an important notification can be predicted with 61% - 90% and 51% - 99% of precision and recall.