Mapping the Hertziana

Hannah Casey, Dario Rodighiero, and Alessandro Adamou

Mapping of the Bibliotheca Hertziana’s collection. The effect of the hyperparameters on the UMAP algorithm illustrated through embeddings using cosine and hamming distances for book loans. The structure of the library appears robust, with a larger cluster of books in the center and smaller ones surrounding it

Mapping of the Bibliotheca Hertziana’s collection. The effect of the hyperparameters on the UMAP algorithm illustrated through embeddings using cosine and hamming distances for book loans. The structure of the library appears robust, with a larger cluster of books in the center and smaller ones surrounding it

Under the supervision of Dr. Alessandro Adamou and Dr. Dario Rodighiero, Hannah Casey led a project to take a different look at the Bibliotheca Hertziana’s classification system and collection. The project used advanced data visualization and machine learning techniques to visually understand the library collection. The primary goal was to develop an analytical method that integrated user loan data with deep mapping techniques. This method sought to graphically display usage patterns within the library, providing a visual classification system. The approach involved UMAP dimensionality reduction to visualize the catalog based on book loans and ChatGPT prompt engineering with large language models (LLMs) to describe loan clusters with detailed summaries and titles. The project's effectiveness was evaluated qualitatively through expert feedback from library stakeholders. Scholars recognized and validated the clusters based on their research areas, confirming that the new mapping system accurately reflected the evolving research themes at Bibliotheca Hertziana. This innovative method offered a scalable solution for visually organizing library collections, enhancing both accessibility and user engagement. It also provided a dynamic framework adaptable to the changing needs of research libraries, potentially applicable to other specialized libraries facing similar classification challenges. Building on the project's success, future research could explore integrating additional data sources like full texts and abstracts to further refine cluster descriptions. There is also potential for developing interactive tools and recommendation systems to enhance user experience and support scholarly research. This project represented a significant step towards modernizing library classification systems, making them more responsive to the needs of contemporary research environments.