Visual Geocoding A Quarter Billion Global News Photographs Using Google's Deep Learning API

Last March I wrote about an early experiment using Google's Cloud Vision API to perform deep learning-powered geocoding of 20 million global news images. In that experiment I compiled 20 million photographs that had appeared in online news articles worldwide as monitored by the open data GDELT Project over a period of two months and ran them through Google's Vision API service, which applies state-of-the-art deep learning algorithms to visually analyze an image much as a human would. The API returns a wealth of data about each image, including a list of objects and activities it depicts, recognizable logos, OCR text recognition in almost 80 languages, levels of violence, an estimate of how "happy" or "sad" people in the photograph appear to be and even the precise location on earth the image appears to depict. It is that last category that is so fascinating when it comes to trying to understand the visual geography of the world's news media. One year later the GDELT Project has now processed more than a quarter billion news photographs from news outlets in almost every corner of the world through Google's API – what can we learn through this deep learning powered "visual geocoding" of the world's news imagery?