Bridging FAIR and CARE in ETD Metadata: An LLM assisted Cross- Repository Evaluation

Abstract

Electronic Theses and dissertations that are submitted electronically are known as electronic theses and dissertations (ETDs), and they are distributed through a variety of institutional and aggregated repositories. Even though the majority of ETD platforms adhere to the FAIR principles, which specify that data must be Findable, Accessible, Interoperable, and Reusable, these platforms frequently fail to take into account the ethical and community-centered aspects that are encapsulated in the CARE principles, which are as follows: Collective Benefit, Authority to Control, Responsibility, and Ethics. The purpose of this research is to present a novel cross-repository evaluation framework that utilises an LLM-assisted technique to bridge the gap between the FAIR and CARE principles. A rubric-based review was combined with the reasoning skills of three big language models - ChatGPT, Grok, and DeepSeek R1 - in order to conduct an evaluation of nine of the important open-access electronic text databases (ETD) repositories. A standardised rubric served as a guide for each model, and it was prompted to conduct an analysis of the quality of the metadata as well as ethical constraints. Despite the fact that the data demonstrate that FAIR compliance is resilient across repositories, they also highlight systemic weaknesses in CARE alignment, particularly with regard to cultural context, ethical reuse, and authorial control. Moreover, the comparative analysis among three agents suggests that it should be used for evaluating FAIR compliance. CARE compliance evaluation may need more sophisticated 'Human in the Loop' setup. This framework offers a scalable and transparent approach to analysing metadata governance. Additionally, it gives schema-agnostic recommendations for encouraging inclusivity and ethical stewardship in digital academic infrastructure. These characteristics are achieved through the triangulation of assessments given by artificial intelligence.