Historical Newspaper Images and Machine-Generated Captions

Over 1.5 million newspaper page images and associated text from the Library of Congress Chronicling America collection. The dataset includes machine-generated annotations for image captioning, text-to-image, and feature extraction tasks. It was created by davanstrien and last updated in October 2022.

Use Cases

• Train image captioning models using the machine-generated text annotations paired with historical newspaper images.
• Perform zero-shot image classification by leveraging the visual features extracted from the page images.
• Conduct text-to-image retrieval experiments using the OCR text associated with each image.
• Analyze feature embeddings generated for the image data for tasks like visual similarity search.
• Study the evolution of newspaper layout and visual content across the temporal range of the collection.

Strengths

• Contains over 1.5 million data points (images and text).
• Multimodal data with image, text, and machine-generated annotations.
• Sourced from the authoritative Library of Congress Chronicling America project.

Limitations

• Machine-generated annotations may contain errors or noise.
• Geographic coverage is primarily focused on the United States (Region:us).
• Data recency is limited, with the last update in 2022.

Provenance

Source

Library of Congress Chronicling America collection.

Collection Method

Machine-generated annotations applied to digitized newspaper pages.

Time Range

Temporal coverage of the historical newspaper collection.

Freshness

Last updated 2022-10-14.

Geography

Primarily United States (Region:us).