Abstract
With the increasing adoption of digitization, more and more historical visualizations created hundreds of years ago are accessible in digital libraries online. It provides a unique opportunity for visualization and history research. Meanwhile, there is no large-scale digital collection dedicated to historical visualizations. The visualizations are scattered in various collections, which hinders retrieval. In this study, we curate the first large-scale dataset dedicated to historical visualizations. Our dataset comprises 13K historical visualization images with corresponding processed metadata from seven digital libraries. In curating the dataset, we propose a workflow to scrape and process heterogeneous metadata. We develop a semi-automatic labeling approach to distinguish visualizations from other artifacts. Our dataset can be accessed with OldVisOnline, a system we have built to browse and label historical visualizations. We discuss our vision of usage scenarios and research opportunities with our dataset, such as textual criticism for historical visualizations. Drawing upon our experience, we summarize recommendations for future efforts to improve our dataset.
Introduction
While iconic works like John Snow’s cholera map or Charles Minard’s flow maps are widely celebrated, they represent only a fraction of the historical visualization record. Millions of other artifacts remain buried within vast digital libraries, largely inaccessible due to fragmentation across institutions, inconsistent metadata schemas, and the sheer volume of non-visual material. To address this "needle in a haystack" problem, we introduce OldVisOnline, the first large-scale dataset specifically dedicated to historical visualizations published before 1950.
Methodology: The Distillation Pipeline
Curating a clean dataset from massive digital archives required a robust, semi-automatic workflow. We initially harvested over 370,000 candidate images from seven major sources, including the British Library, the Library of Congress, and the Internet Archive. To normalize the heterogeneous metadata provided by these institutions, we developed a unified schema covering 17 dimensions, including high-resolution URLs and perceptual hashes for duplicate detection.
To isolate visualizations from text-heavy pages and conventional geographic maps, we implemented a three-stage classification pipeline using VGG-16 binary classifiers:
- Text Filtering: Removing pages consisting solely of text (approximately 45% of the initial pool).
- Map vs. Visualization: Distinguishing between standard geographic maps and non-spatial data visualizations.
- Refinement: Isolating pure visualizations from complex, hybrid map-based artifacts.
Following automated classification, we conducted expert quality assurance to correct misclassifications, resulting in a final dataset of 13,511 high-quality visualization images.
The OldVis Dataset and System
The OldVis dataset represents a diverse cross-section of historical data design, featuring metadata in 38 languages. Analysis of the metadata reveals significant historical trends, such as the professionalization of the field in the late 19th century through prolific authors like Henry Gannett. To facilitate exploration, we developed the OldVisOnline system, which allows researchers to browse, search, and further label the collection.
Research Opportunities: Textual Criticism
A primary application of this dataset is "textual criticism"—the comparative study of variations between different editions of the same visualization. For instance, our analysis of John Snow’s cholera map revealed that a 1936 reprint contained a printing defect where ink diffusion merged two distinct cholera cases into one. Such variations can lead to significantly different modern data reconstructions and interpretations.
Beyond textual criticism, the dataset serves as a repository of design inspiration. Historical elements like "volvelles" (interactive rotating paper wheels) offer unique precedents for modern interactive and physical visualization designs.
Conclusion and Future Work
OldVisOnline provides a foundational resource for the visualization community, though it is subject to the "survival bias" of what has been digitized to date. Future efforts will focus on integrating Optical Character Recognition (OCR) for semantic searching of chart contents and implementing more fine-grained labeling of chart types (e.g., distinguishing bar charts from line graphs). We invite the community to contribute to this growing archive at github.com/oldvis.
Summary updated