Get started
Work

Machine Learning for Metadata Creation in Libraries : An Exploratory Study of AI Tools for Subject Cataloguing and Classification

Year: 2025
Type: dissertation
Abstract: This study explores the potential of large language models (LLMs) to support subject cataloguing and classification in libraries by evaluating two LLM tools—ChatGPT-4o and DeepSeek—against professionally catalogued records, and reflects on their possible role in cataloguing practice. Using a sample of 60 bibliographic records from three subject domains, 20 from each of Physics, Plant Culture, and Decorative Arts, the study compares AI-generated metadata with human-assigned Library of Congress Subject Headings and Library of Congress Classification codes. A mixed-methods approach was applied, combining precision/recall-based metrics with expert qualitative evaluations applied to a subset of 25 records. The results show that while AI tools can generate thematically appropriate subject headings and plausible classification codes, exact matches with human metadata remain limited. ChatGPT-4o was more consistent in formatting and use of authorized terms, whereas DeepSeek produced more varied and occasionally more specific suggestions but with less structural control. Performance ranged from a Partial F1 of 0.42 to Full F1 scores of 0.28 with ChatGPT-4o and 0.26 with DeepSeek. Subject area influenced these outcomes, with structured fields such as Physics supporting stronger results, while more interpretive or applied domains like Decorative Arts and Plant Culture contributed to lower scores. The findings suggest that while LLMs show promise for supporting cataloguing workflows, human oversight remains essential, especially in nuanced or domain-specific contexts where accurate subject assignment often depends on specialized knowledge.
Author: Osmani, Ilektra
Language: English
Open Access status: green
OpenAlex ID: W7112508473
Cites: 31
Cited by:
Related to: 10