Large-Scale Sleep Condition Analysis Using Selfies from Social Media

Sleep condition is closely related to an individual's health. Poor sleep conditions such as sleep disorder and sleep deprivation affect one's daily performance, and may also cause many chronic diseases. Many efforts have been devoted to monitoring people's sleep conditions. However, traditional methodologies require sophisticated equipment and consume a significant amount of time. In this paper, we attempt to develop a novel way to predict individual's sleep condition via scrutinizing facial cues as doctors would. Rather than measuring the sleep condition directly, we measure the sleep-deprived fatigue which indirectly reflects the sleep condition. Our method can predict a sleep-deprived fatigue rate based on a selfie provided by a subject. This rate is used to indicate the sleep condition. To gain deeper insights of human sleep conditions, we collected around 100,000 faces from selfies posted on Twitter and Instagram, and identified their age, gender, and race using automatic algorithms. Next, we investigated the sleep condition distributions with respect to age, gender, and race. Our study suggests among the age groups, fatigue percentage of the 0-20 youth and adolescent group is the highest, implying that poor sleep condition is more prevalent in this age group. For gender, the fatigue percentage of females is higher than that of males, implying that more females are suffering from sleep issues than males. Among ethnic groups, the fatigue percentage in Caucasian is the highest followed by Asian and African American.