Files in this item



application/pdfSONG-THESIS-2021.pdf (20MB)Restricted to U of Illinois
(no description provided)PDF


Title:Investigating representational biases in stock photo website search results
Author(s):Song, Hang
Advisor(s):Karahalios, Karrie
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Stock photo
algorithm auditing
social computing
Abstract:Stock photos have been widely used in various media, such as newspapers and blogs. They are relatively understudied in media and advertising, perhaps because the concept of stock photos encompasses a wide range of images, and websites dedicated to gathering, displaying, and selling stock photos haven't been popular until recent decades. Therefore, this paper aims to bring into light the representation biases that digital stock photos have regarding races and genders across different occupations. Compared to the employment statistics dataset published by the US Labor of Bureau Statistics and to the ideal situation where all genders and races are represented equally, stock photos collected from Shutterstock have demonstrated different degrees of representation bias regarding perceived race and gender. Stock photo creators have used several techniques to make their images suitable for a wide range of topics and audiences, including using illustrations instead of actual photos, taking pictures about parts of human bodies (primarily hands and arms), and depicting people's silhouettes and shadows. These techniques can undoubtedly improve the diversity of stock photos, but their effects are limited as many visual cues for genders and races still exist. Finally, whether jobs are popular in rural and metropolitan areas in the United States has mixed effects on perceived gender and racial biases.
Issue Date:2021-04-28
Rights Information:Copyright 2021 Hang Song
Date Available in IDEALS:2021-09-17
Date Deposited:2021-05

This item appears in the following Collection(s)

Item Statistics