TextMachineProject/blbooks_metadata

19th Century Digitised Books Collection: BLBooks Metadata

Dataset Summary

This dataset consists of structured metadata from the British Library 19th Century Digitised Books Collection (henceforth blbooks), a large collection of digitized books in the public domain. It has been created from the OCR derived text + metadata version (henceforth blbooks_derived) of the original dataset, which is released under CC0 1.0 Universal Public Domain license. The complete implementation code is available at: https://github.com/Text-Machine/data-processing-code. We have processed the records to include a genre label (fiction or non-fiction) assigned by Gemma4-31b-it based on the title and the author. We also used the dedupe library to perform fuzzy string matching for duplicate detection. All duplicated records are grouped under the same book_id, and the one with the highest OCR quality is flagged as best_version.

Languages and Periods

Records have been filtered to include only books tagged as written entirely in one of following languages: English, French, Spanish, Italian, Dutch, Russian.

Language	Number of records
English	36924
French	3734
Spanish	742
Italian	643
Russian	564
Dutch	535

Records per period (respecting original period split in blbooks_derived):

Period	Number of records
1510_1699	680
1700_1799	1782
1800_1809	1062
1810_1819	1870
1820_1829	1847
1830_1839	1873
1840_1849	2815
1850_1859	3923
1860_1869	4439
1870_1879	5502
1880_1889	7248
1890_1899	10101

The resulting metadata file consists of 43,142 entries.

Structure

The dataset is provided as a tabular file with the following columns:

Column	Description
record_id	The unique identifier (from blbooks_derived, field record_id).
date	The publication year of this specific edition (from blbooks_derived, field date).
title	The full title of the work (from blbooks_derived, field title).
place	The place of publication (from blbooks_derived, field place).
mean_wc_ocr	Word OCR quality averaged per record (from blbooks_derived, obtained by averaging the per-page mean word confidence values, field mean_wc_ocr).
std_wc_ocr	Standard deviation of the word confidence values averaged per record (from blbooks_derived, obtained by averaging the per-page standard deviation values, field std_wc_ocr).
author	Main author of the publication (from blbooks_derived, field Name).
all_authors	All names associated with the publication (from blbooks_derived, field all names).
publisher	Publisher of the work (from blbooks_derived, field Publisher).
main_publication_country	First country associated with publication (from blbooks_derived, field Country of publication 1).
all_publication_countries	A list of all associated countries of publication (from blbooks_derived, field all Countries of publication).
main_language	The primary language associated with the book (from blbooks_derived, field Language_1).
number_of_pages	Total page count.
number_of_words	Total word count (using .split()).
path_to_json	File path to the corresponding raw JSONL content (pointing to the corresponding content file in blbooks_derived, where each row corresponds with one page of the book).
book_id	Unique identifier for the work after deduplication.
best_version	Boolean: 1 if this is the highest quality OCR version, 0 otherwise.
genre	Genre classification (fiction or non-fiction) assigned by Gemma 4.

Data instance

An example data instance:

{
    'record_id': 3112351,
    'date': 1692,
    'title': 'The Traytor, a tragedy [in five acts and in verse] with alterations, amendments and additions. Written by Mr. Rivers',
    'place': 'London',
    'mean_wc_ocr': 0.4944406779661017,
    'std_wc_ocr': 0.1780338983050847,
    'author': nan,
    'all_authors': 'Shirley, James, 1596-1666 [person]',
    'publisher': nan,
    'main_publication_country': 'England',
    'all_publication_countries': 'England',
    'main_language': 'English',
    'number_of_pages': 59,
    'number_of_words': 18462,
    'path_to_json': '1510_1699/1692_003112351.jsonl.gz',
    'book_id': 37591,
    'best_version': 1,
    'genre': 'fiction'
}

Duplicates and best version

An example of deduplication with best version selection:

import pandas as pd
df = pd.read_csv("metadata_blbooks.csv")
df[df["author"] == "Austen, Jane"][["title", "author", "date", "mean_wc_ocr", "book_id", "best_version", "genre"]]

record_id	title	author	date	mean_wc_ocr	book_id	best_version	genre
144506	Emma : a novel	Austen, Jane	1870	0.830995	1195	1	fiction
144565	Mansfield Park : a novel	Austen, Jane	1875	0.706090	1192	0	fiction
144563	Mansfield Park : a novel	Austen, Jane	1870	0.850376	1192	1	fiction
144766	Sense and sensibility	Austen, Jane	1884	0.569446	1193	0	fiction
144597	Northanger Abbey : and, Persuasion	Austen, Jane	1897	0.810952	1197	1	fiction
144469	The novels of Jane Austen	Austen, Jane	1898	0.861404	1196	1	fiction
144559	Mansfield Park, etc	Austen, Jane	1898	0.788097	1192	0	fiction
144749	Sense and sensibility	Austen, Jane	1896	0.816648	1193	1	fiction
144509	Emma	Austen, Jane	1896	0.772656	1195	0	fiction
144558	Mansfield Park	Austen, Jane	1897	0.830825	1192	0	fiction
144653	Pride and prejudice	Austen, Jane	1895	0.823202	1194	1	fiction

Additional information

For additional information on the original dataset, read the dataset card at https://huggingface.co/datasets/TheBritishLibrary/blbooks.