Exploring Computational Descriptions for Metadata Creation for E-Books at the Library of Congress, United States of America

The Library of Congress’ (LoC) project “Exploring Computational Description” (ECD) is investigating the use of machine learning (ML) to create metadata for e-books that have not yet been catalogued. In-house the LC Labs carried out this initiative with the U.S. Programs, Law, and Literature Division and an external vendor. An initial budget of $250,000 from the National Digital Trust Fund was allocated for this experimental AI endeavour, prompted by a massive backlog of e-books. During the first project phase, five ML models were evaluated, and in the second project phase, human-in-the-loop prototypes that offer machine-generated terms to librarians were introduced. The integration of AI at the LoC has the potential to enhance cataloguing efficiency by automating repetitive tasks, thereby allowing librarians to focus more on intellectual tasks. At the same time, the project faced several challenges, including ensuring the reliability of AI-generated records, copyright concerns, and managing potentially harmful language in older texts used for training the models. Improving the accuracy of these models remains essential and depends on access to extensive digital data. However, human expertise remains crucial for ensuring high quality, and librarians need to develop a foundational understanding of ML to leverage these technologies effectively. The aim of the project is to develop innovative approaches that contribute to improving library practices.