Position: Measure Dataset Diversity, Don't Just Claim It
This work addresses the issue for ML researchers and dataset curators by advocating for more precise handling of value-laden properties, though it is incremental in applying existing social science principles to ML.
The paper tackles the problem of vague and unvalidated terms like 'diversity' in machine learning datasets by analyzing 135 image and text datasets, applying measurement theory from social sciences to provide recommendations for conceptualizing and evaluating diversity.
Machine learning (ML) datasets, often perceived as neutral, inherently encapsulate abstract and disputed social constructs. Dataset curators frequently employ value-laden terms such as diversity, bias, and quality to characterize datasets. Despite their prevalence, these terms lack clear definitions and validation. Our research explores the implications of this issue by analyzing "diversity" across 135 image and text datasets. Drawing from social sciences, we apply principles from measurement theory to identify considerations and offer recommendations for conceptualizing, operationalizing, and evaluating diversity in datasets. Our findings have broader implications for ML research, advocating for a more nuanced and precise approach to handling value-laden properties in dataset construction.