Evaluation of AI-Generated Keywords for Information Retrieval in Library Catalogues

Abstract

Artificial Intelligence (AI) has been receiving increasing attention in recent years. Yet its practical implementation remains somewhat limited in libraries. Despite notable advancements, several critical aspects of library technical work, particularly classification and cataloguing, continue to rely heavily on manual processes. This paper examines challenges in current practices, notably the labour-intensive tasks of finding class numbers and subject headings. While the option to import MARC (machine-readable cataloging) records via protocols like Z39.50 provides some relief, the absence of readily available records presents significant obstacles, especially for locally published materials. Against this backdrop, this research explores AI’s potential in generating relevant subject headings to streamline cataloguing processes and augment information retrieval. This study analysed one hundred books from the Indian Institute for Human Settlements (IIHS) library, employing ChatGPT to generate subject headings. However, the resulting headings exhibited a lack of uniformity compared to manual cataloguing. Library headings emphasized geographic and physical details, whereas ChatGPT-generated headings were less detailed. Although ChatGPT facilitates the instantaneous generation, the quality of headings varied, with some resembling MARC fields. Refinement of ChatGPT’s syntax could potentially enhance relevance. Despite its efficiency, ChatGPT cannot entirely replace the human judgment required in cataloguing processes.