Towards Measuring Fine-Grained Diversity Using Social Media Photographs

Diversity is an important socio-economic construct that influences multiple aspects of human lives from the prosperity of a city to corporate earnings and from criminal justice to health and social engagement. Large, heavily populated urban areas can be highly diverse at the city or even neighborhood level, but we know very little about how much people from diverse demographics (such as age and race) interact with each other. Previous work has shown that photos are important in social relationships. The growing presence of photos online and on social media, therefore presents a unique opportunity to study diversity in interactions. In this paper, we explore a novel approach to measure p-diversity i.e. a personal, photo-level diversity metric computed using social media data. Specifically, we focus on Instagram photos of multiple people interacting, and employ automatic methods for race, age, and gender estimation to quantify mixing in such photos. We compare and contrast this new measure of diversity with traditional (i.e. census-based) metrics using a dataset for New York City. Results obtained motivate the use of social media photos to complement census data to develop cheaper, faster, mechanisms for studying diversity and applying them in social, economic, political, and urban planning contexts. Diversity is an important socio-economic construct that has associations with multiple aspects of human life including commerce, innovation, well-being, criminal justice, civic responsibilities and health among others (Page 2007). Traditionally, diversity has been defined as a function of the number of people of different age, gender, ethnicity etc. living in the same neighborhood as observed through long-term data e.g. census (Maly 2000). However, there exist multiple nuances in the notion of diversity, some of which remain hidden if we work with coarse, long-term, residential measures of diversity. For example, zooming further into the location aspect, it has been reported that the areas which appear to be most diverse at city scale are often also the most segregated, when observed at a neighborhood scale (Silver 2015). Census data (such as from the American Community Survey) provide an important public service by collecting and preserving demographics data going back for decades. Despite its inherent value, census based data collection and analysis has several shortcomings including being expensive, time consuming and labor-intensive. In much previous Copyright c © 2017, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. work, it has also been argued that studying informal interactions is important to understand social diversity, but it was hard to study them all along (e.g. (Hays and Kogl 2007)). There is hence much motivation to complement traditional census metrics with a novel methodology that leverages the increasing amount of information that is available from new sources of data such as social networks. Previous work has shown that photos are important in social relationships. The content of photos shows who is part of a group and telling stories about photos helps nurture relationships (Van House et al. 2005). Hence, here we assume that people sharing the same “frame” in a single photo have some kind of interconnection among themselves and proceed to define a new fine grained metric for measuring diversity. While similar in spirit to the traditional measures of diversity, the proposed p-diversity metric quantifies the level intermixing occurring between people of different demographic descriptions in their personal spaces as observed via social media photographs. The proposed p-diversity metric allows the notion of diversity to vary in (near) real-time rather than assumed to be a constant between census updates and focuses on diversity of relationships between people rather than their residential addresses. In this short article, we describe a methodology to compute this new p-diversity metric using social media (Instagram) data and undertake a case-study analysis by comparing the results obtained by the proposed metric with those obtained via traditional census based metrics in New York City. The obtained results motivate and ground the use of social media photographs for studying diversity.