Exploring the Use of Machine Vision to Understand Neighborhood Inequality

• In Event: 2204 - New Approaches to Understanding and Addressing Inequality

Abstract

The online world has become one of the primary ways people construct, present, and reconstruct their identities, and images are rapidly becoming the main currency in the online world. While social scientists have successfully applied text analysis tools built by computer and information scientists to questions important to the social sciences – known as the text-as-data field – social scientists have yet to do the same with images-as-data. As the use of images on social media proliferates, the content of images constitute a growing and mostly untapped source of knowledge about our social world. In particular, as we increasingly view each other’s lives through GIF-ified emotions, filters, and images, it is important to understand whether online images reflect, reinforce, or have the power to challenge existing group-based inequalities. In this paper we ask two broad questions: (1) Methodologically, can we use machine-vision tools, at scale, to extract data from the content of images relevant to social science inquiry? (2) Substantively, is online image sharing bridging inter-group inequalities, or reinforcing them? To answer the first question, we use a two step approach based on the computational grounded theory framework. We first cluster images using Principle Component Analysis and hierarchical clustering, inductively analyzing the content of each cluster; and second, we use object detection algorithms to quantitatively analyze clusters in more detail. We explore the second, substantive, question by using the above methods to analyze gentrification, neighborhood stigma, and the potential for marginalized communities to use online spaces to reclaim their own representations. The answers to these questions, we conclude, have implications for how we understand neighborhood, community, and intra-group inequalities in contemporary society.

Authors

• Laura K. Nelson, Northeastern University

• Jeffrey L. Sternberg, Northeastern University

• Tirthak Lalitbhai Patel, Northeastern University

• Devesh Tiwari, Northeastern University