Detecting and Predicting Beautiful Sunsets Using Social Media Data

Emma Pierson, Stanford University

Beautiful sunsets are one of the few pleasures freely available to everyone, but due to their subjective nature are difficult to quantitatively study on a large scale. Here, we use 1.2 million sunset posts on Instagram, a picture-sharing platform, to detect beautiful sunsets in 10 American cities over 7 months. We show that our metric of sunset quality correlates with human assessments, make sunset quality scores publicly available to allow more systematic study of sunsets, and use this dataset to answer a number of basic questions. Do some locations have more beautiful sunsets than others? Are there meteorological features which predict beautiful sunsets? Does a beautiful sunset today predict a beautiful sunset tomorrow? Is it possible to detect beautiful sunsets early enough to notify people so they can go outside to enjoy them? What visual features are people responding to when they call a sunset beautiful? We validate a widely used sunset prediction model developed by meteorologists, produce an algorithm which can visually discriminate between beautiful and mediocre sunsets, and provide a messaging service and web interface to notify users of beautiful sunsets.