Existing sleep-tracking apps and devices provide simple descriptive statistics or generic recommendations for everyone. In this work, we aim to leverage cohort-based sleep data to provide recommendations to improve an individual's sleep. We report a 4-week study (N = 39) conducted to compare three alternatives: 1) no recommendation, 2) general recommendation, and 3) cohort-based recommendation, using six sleep quality metrics. For the cohort-based recommendation, recommendations were generated based on "similar users" using about 40 million sleep events from Microsoft Band users. Our results indicate that cohort-based systems for health recommendations can prompt a desire for behavior change inspired by social comparison and increased awareness about sleep habits. However, in order to be effective, such systems need to establish their credibility and to be able to generate cohorts based on features that are important to users. Finally, we provide further suggestions and design implications for future cohort-based recommendation systems for healthy sleep.

No abstract

No abstract