From Open‑Ended Text to Taxonomy: An LLM‑based Framework for Information Sources for Disability Services

Abstract

People with disabilities (PWD) and their family members often find it difficult to find information about available services. One of the approaches to address this information access problem is by understanding the ecology of available information sources. However, identifying the landscape of information sources is challenging due to the variety of sources and their varying visibility. This study proposes a computational approach to processing open-ended survey answers by constructing a hierarchical taxonomy of information sources. We developed a semi-automated, LLM-based framework to build a taxonomy of information sources from open-ended survey answers. The resulting 3-tier taxonomy captures broad categories and fine-grained entities, supporting multi-level analysis of information sources. This work explores the feasibility of LLM-based taxonomy building and offers a scalable framework for processing open-ended texts.