Designing notifications in Augmented Reality (AR) that are noticeable yet unobtrusive is challenging since achieving this balance heavily depends on the user’s context. However, current AR systems tend to be context-agnostic and require explicit feedback to determine whether a user has noticed a notification. This limitation restricts AR systems from providing timely notifications that are integrated with users’ activities. To address this challenge, we studied how sensors can infer users’ detection of notifications while they work in an office setting. We collected 98 hours of data from 12 users, including their gaze, head position, computer interactions, and engagement levels. Our findings showed that combining gaze and engagement data most accurately classified noticeability (AUC = 0.81). Even without engagement data, the accuracy was still high (AUC = 0.76). Our study also examines time windowing methods and compares general and personalized models.
https://dl.acm.org/doi/10.1145/3706598.3713511
The ACM CHI Conference on Human Factors in Computing Systems (https://chi2025.acm.org/)