Generative AI-augmented and User-centric Research Data Discovery and Reuse

Abstract

This dissertation addresses the challenge of enhancing research data discovery and reuse in the face of escalating data volume and complexity. Traditional metadata-driven search tools often fall short in providing nuanced context and interdisciplinary connections critical for efficient scientific exploration and collaboration. To address these limitations, we developed the Generative AI-augmented and User-centric Data Search (GAUDS) system, which integrates Large Language Models (LLMs) and Scholarly Knowledge Graphs (SKGs) to parse natural language queries and visualize data relationships, thereby fostering a deeper understanding of available research resources.

The study details the development and implementation of the GAUDS system, including the conceptualization of the guiding principles Connectivity, Effectiveness, Visibility and Interactivity (CEVI) that support and evaluate the discovery and reuse of research data. It further explores the construction of the ICPSR Health and Medical Scholarly Knowledge Graph (IHSKG), which represents complex connections in research data and prototypes interdisciplinary reuse potentials. The abilities of LLMs to perform complex reasoning were assessed, informing the system's ability to understand and manipulate large datasets effectively.

The development of the GAUDS system, informed by insights gained from prototyping and evaluating user-centric utility, leads to a comprehensive analysis of focus group feedback. This feedback evaluates the system’s impact on enhancing data discoverability and usability. The GAUDS system, by providing effective navigation aids, relevant dataset suggestions, and contextualized reuse guides, not only enhances user engagement and satisfaction, but also demonstrates the transformative potential of generative AI in specialized academic domains such as health and medical research. This research contributes to the fields of information retrieval and data management by proposing a novel approach that combines human-curated knowledge graphs with generative AI algorithms to significantly improve data discovery and reuse. Future work will aim to productionize the GAUDS system, expand its scalability across different domains, and explore its broader potential to support open science initiatives.