No abstract

No abstract

The visual portrayal of social groups in media reinforces stereotypes and narratives, potentially leading to discriminatory actions and policies. That is particularly true for underrepresented or stigmatized groups such as migrants and is a phenomenon that varies per country. Therefore, studying the representation of migrants requires analyzing considerable amounts of visual data from different locations. This work addresses that challenge with an interdisciplinary approach characterizing the visual portrayal of migrants using Deep Learning techniques and analyzing results through the lenses of migration and gender studies. Images associated with migrants found on the internet through a search engine and from ten countries are processed to quantify and analyze the demographic and emotional information of the people portrayed. An intersectional approach is employed regarding gender, age, physical features, and emotions. The general group “migrants” is compared with the specific groups “refugees” and “expats”. Results suggest that portrayals predominantly focus on asylum seekers and associate them with poverty and risks for host societies. Moreover, the demographics in the portrayals do not match the official statistics. For expats, an over-representation of “white” and an under-representation of “asian” faces were found, while for migrants and refugees, depictions align with the demographics of low-skilled migrants. Furthermore, results evidence the power struggle underlying the “expat vs. migrant” dichotomy and its inherent colonial nature. The emotions displayed are predominantly negative and align with emotional and gender stereotypes literature. Positive emotions are more associated with women than men, and with expats than refugees and migrants. Previous results regarding the under-representation of migrant women in media are confirmed. Also, women are portrayed as younger than men, and expat women are the youngest. Children appear more in pictures associated with refugees and migrants than with expats. Likewise, migrants are often depicted as crowds, but when that is not the case, migrant and refugee women appear in larger groups than men. A higher proportion of images associated with expats do not contain people. All these effects, however, differ per location. Finally, we suggest future directions and analyze possible limitations of automatic visual content analysis using existing Deep Learning models.

No abstract