svattikuti/rev_warwithembeddings

Revolutionary War With Embeddings

This dataset repo aggregates several Revolutionary War era corpora and derived embedding tables in Parquet form. It is intended for retrieval, search, clustering, OCR exploration, and multimodal linking across archival text and images.

The repository currently contains 21 files, primarily Parquet datasets, plus a few small CSV/JSON artifacts used during processing.

What is included

The dataset combines material from three main source families:

• Library of Congress Chronicling America newspaper data and OCR.
• U.S. National Archives Revolutionary War pension and related records.
• Smithsonian Institution Revolutionary Era collection and media data.

It also includes derived embedding tables:

• *_e5_embeddings.parquet files for text embeddings.
• si_us_revolutionary_era_media_clip_embeddings.parquet and si_us_revolutionary_era_media_clip_embeddings_with_context.parquet for image or image-context CLIP embeddings.

Key files

Some of the larger and more central files are:

• loc_chronicling_america_1770_1810.parquet: 58,116 rows of newspaper/OCR-oriented records.
• nara_revolutionary_war_pension_files.parquet: 2,244,629 rows of NARA pension-file related records.
• si_us_revolutionary_era_collections.parquet: 12,667 Smithsonian collection records.
• si_us_revolutionary_era_media_blobs.parquet: 5,205 media records with binary image payloads.

Example schemas

loc_chronicling_america_1770_1810.parquet includes fields such as:

• Web_URL
• newspaper_title
• place_of_publication
• issue_date
• Page
• thumbnail_url
• jpeg2000_url
• pdf_url
• ocr_url
• ocr_text

nara_revolutionary_war_pension_files.parquet includes fields such as:

• NAID
• naraURL
• title
• logicalDate
• pdfURL
• pageURL
• extractedText
• transcriptionText
• transcriptionUserNames

si_us_revolutionary_era_media_blobs.parquet includes:

• media_key
• media_url
• image_bytes
• size_bytes

Intended use

Possible use cases:

• Semantic search over Revolutionary War newspapers and archival text.
• Linking text records to embeddings for retrieval-augmented workflows.
• Matching Smithsonian media assets to text context using CLIP embeddings.
• OCR quality inspection and downstream historical NLP experiments.
• Building exploratory knowledge bases around Revolutionary War people, places, and events.

Notes

• Files are stored as flat artifacts rather than as named Hugging Face splits.
• Several files are derivative products generated from source records, including embeddings and image blobs.
• The .parquet files are the primary dataset payloads; the .csv and .json files are smaller supporting outputs.

Loading examples

Python with pandas

import pandas as pd

df = pd.read_parquet("hf://datasets/svattikuti/rev_warwithembeddings/loc_chronicling_america_1770_1810.parquet")
print(df.head())

DuckDB

SELECT *
FROM read_parquet('hf://datasets/svattikuti/rev_warwithembeddings/nara_revolutionary_war_pension_files.parquet')
LIMIT 10;

Limitations

• OCR text may contain recognition errors.
• Coverage and row granularity differ across source families.
• Embedding files depend on the model and preprocessing choices used during dataset construction.
• Some tables contain binary blobs and may be large to download or query locally.

Provenance

Source provenance is inferred from file names and record schemas in this repository:

• Library of Congress / Chronicling America
• U.S. National Archives and Records Administration
• Smithsonian Institution collections and media

If stricter citation, licensing, or collection-level provenance is needed, those details should be added in a follow-up update.