Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
Please note that all times are shown in the time zone of the conference. The current conference time is: 14th June 2025, 06:08:24pm WEST
Find everyone? Scaling up scanned document automated processing of millions of census records to reconstitute the French population in the Socface project
Christopher Kermorvant1, Lionel Kesztenbaum2,3, Yannick Dupraz4,2
1TEKLIA, France; 2INED - Institut national d'études démographiques; 3PSE - Paris School of Economics; 4Université Paris Dauphine-PSL
The Socface project aims to process the complete French historical censuses (1836-1936) using integrated handwriting recognition models to manage millions of records. Challenges include scaling workflows, ensuring data quality and developing integrated models for different layouts. The entire database will be openly accessible to enable extensive social and historical research.
Enhancing Text-to-Image Alignment with Retrieval-Augmented GPT for Historical Event Reconstruction: Evaluating with Multimodal LLMs
Zejie Guo, Phillip Benjamin Ströbel, Felix Klaus Maier
University of Zurich, Switzerland
Enhancing text-to-image (T2I) models for historical event reconstruction involves refining prompts with retrieved context via GPT-4o. This study evaluates alignment using QG-VQA metrics and Likert-scale ratings with history students and MLLMs. Results show improved performance on DALL-E 3, FLUX.1, and SDXL, surpassing baseline models and human-generated prompts.
Illustrated Ideologies: A Scalable Viewing of Visual Media in German Children’s Books of the long 19th century
Manuel Burghardt1, Janos Borst1, Wiebke Helm2
1Computational Humanities Group, Leipzig University; 2Primary School Didactics, Leipzig University
This paper explores the visual dimension of German children’s literature (1801–1914) using a scalable viewing approach. By combining deep learning models and exploratory tools, we analyze 230,000 illustrations to uncover patterns in reading, play, and teaching scenes. The method bridges close and distant viewing, enriching research on historical visual archives.
