IncluSet: A Data Surfacing Repository for Accessibility Datasets.

Area of Science:

• Computer Science
• Artificial Intelligence
• Disability Studies

Background:

• Data sharing is crucial for AI innovation, benchmarking, bias mitigation, and understanding complex AI applications.
• A significant gap exists in available datasets generated by individuals with disabilities, hindering AI model development and evaluation.
• Challenges include smaller population sizes, diverse characteristics, data annotation expertise, and privacy concerns, making data collection and access difficult.

Purpose of the Study:

• To introduce IncluSet, a novel data surfacing repository designed to facilitate the discovery and linking of accessibility datasets.
• To address the scarcity of data from people with disabilities for training and evaluating machine learning models.
• To improve the discoverability of existing and new accessibility datasets for researchers and the disability community.

Main Methods:

• Developed IncluSet, a specialized repository for accessibility datasets.
• Pre-populated IncluSet with information on 139 datasets (65 public, 25 available upon request, 49 not shared but described).
• Designed IncluSet for broad discoverability, including integration with Google Dataset Search.

Main Results:

• IncluSet currently catalogs 139 accessibility datasets.
• The repository includes datasets that are publicly available, available upon request, and those described in literature but not shared.
• IncluSet is engineered to enhance the visibility of these crucial datasets.

Conclusions:

• IncluSet provides a centralized platform for discovering and linking accessibility datasets.
• The repository aims to mitigate the data scarcity challenge for AI development involving people with disabilities.
• By improving dataset discoverability, IncluSet supports innovation and reduces bias in AI applications.