Marching with the Pink Parade: Evaluating Visual Search Recommendations for Non-binary Clothing Items

TL;DR

This study evaluates the poor performance of visual search recommendation systems for non-binary clothing items on e-commerce platforms, highlighting inclusivity issues and suggesting the need for more inclusive algorithms.

Contribution

It provides a comprehensive analysis of how current visual search systems underperform for non-binary clothing, revealing biases and advocating for more inclusive e-commerce recommendations.

Findings

Visual search is less effective for non-binary clothing items.

Participants rate non-binary recommendations lower than binary ones.

Male raters tend to make more binary judgments.

Abstract

Fashion, a highly subjective topic is interpreted differently by all individuals. E-commerce platforms, despite these diverse requirements, tend to cater to the average buyer instead of focusing on edge cases like non-binary shoppers. This case study, through participant surveys, shows that visual search on e-commerce platforms like Amazon, Beagle.Vision and Lykdat, is particularly poor for non-binary clothing items. Our comprehensive quantitative analysis shows that these platforms are more robust to binary clothing inputs. The non-binary clothing items are recommended in a haphazard manner, as observed through negative correlation coefficients of the ranking order. The participants also rate the non-binary recommendations lower than the binary ones. Another intriguing observation is that male raters are more inclined to make binary judgements compared to female raters. Thus it is clear that these systems are not inclusive to the minority, disadvantaged communities of society, like LGBTQ+ people. We conclude with a call to action for the e-commerce platforms to take cognizance of our results and be more inclusive.