From online searches to social media posts, our everyday interactions with the Internet are creating vast amounts of data. Large volumes of this data can be accessed rapidly at low cost, opening up unprecedented possibilities to monitor and analyse social processes and measure human behaviour.
As Internet connectivity has continued to improve, photo-sharing platforms such as Instagram and Flickr have gained widespread popularity. At the same time, considerable advances have been achieved in the power of computers to analyse the contents of images. In particular, deep learning based methods such as convolutional neural networks have radically transformed the performance of systems seeking to identify objects in images, or classify the contents of a scene.
Here, we showcase a series of studies in which we seek to quantify various aspects of human behaviour by exploiting both the large quantities of photographic data shared online and recent developments in computer vision. Specifically, we investigate whether data extracted from photographs shared on Flickr and Instagram can help us track global protest outbreaks; estimate the income of inhabitants living in different areas of London and New York; and predict the occurrence of noise complaints in New York City.
Our findings are in line with the striking hypothesis that information extracted through automatic analysis of photographs shared online may help us measure human behaviour, whether in individual cities or across the globe