Opportunistic Sensing of Behavioural Data to Prioritise Smartphone Notifications in the Wild

Catching the attention of users, via their smartphone, is most often achieved through push notifications. Such an approach is already in place within applications such as Facebook, Google, Instagram, TikTok, and Twitter. Emails and messaging apps are also used to send notifications, which can assist users who may otherwise miss important events. As a result of these multiple services the user may be overwhelmed with notifications and frequently interrupted during key task. To address this concern, a mobile app has been developed for collecting different phone events and sensors data “in-the-wild”. This data which is then used as the basis for behavioural modeling to profile a user and subsequently detect the best time to send a notification. Within the mobile app, a service is running continuously in the background, collecting sensor data (such as accelerometer and gyroscope data), notification data such as notification title, and user response data relating to engagement with notifications and phone events such as foreground app name, battery percentage etc. Additionally, the app collects user cognitive state data through self-reporting. This paper analyse a dataset contributed by five users. In total, 2766 notifications were received over period of up to 10 days. 78% percent of the time, users ignored or remove notifications based on the data analysis. It is therefore a serious issue in the realm of ubiquitous computing that there is a constant bombardment of push notifications.