Who’s Coming? An AI-powered Visual Analysis of the Refugee Crisis Media Coverage

• In Event: Ingroups and Outgroups in the News

Abstract

The media bring the world to our fingertips, and in 2015 media reports helped citizens across the world live the massive exodus of Syrian refugees. Images are processed much quicker than words, making them an element that both casual and in-depth online readers are exposed to. Simple exposure to images has been shown to activate emotions and stereotypes. This paper relies on AI visual recognition advances to analyze the visual representation of the refugees’ arrival in Europe at the peak of the crisis, in August and September 2015. We use Mask-RCNN and Face Landmark Estimation to analyze the content of the visuals, and KH Coder 3 to analyze the image captions. Overall, we examine over 1500 images that appeared in the top online news sources over 4 weeks in the UK (bbc.com) and France (lefigaro.fr and lemonde.fr). By tracking the number of people depicted in images, the visibility of their faces, and the sentiment expresses in the image captions, we are able to provide a comparative analysis of the visual coverage on a daily basis across countries and across outlets on different sides of the ideological spectrum. Contrary to expectations, we find that rather than large groups, those arriving in Europe were portrayed in medium sized groups between 5-10 people on average. Moreover, we find that the tone of the image caption (positive or negative) is only mildly related to the visual portrayals (specifically to the number of people in an image and to the extent their face is visible). We find significant differences when comparing left and right-wing news outlets’ portrayals in France. This paper highlights the importance of AI-assisted visual analysis for a better understanding of visual bias in public news.

Author

• Delia Dumitrescu, University of East Anglia