license: cc-by-4.0
task_categories:
- text-classification
- feature-extraction
language:
- de
- la
- en
- fr
- it
tags:
- metadata
- cultural heritage
- research library
- library sciences
- machine learning
- data science
pretty_name: ark-metadata
size_categories:
- 1M<n<10M
Metadata of the "Alter Realkatalog" (ARK) of Berlin State Library (SBB)
Motivation
This dataset was created with the intent to provide a single larger set of metadata from Berlin State Library for research purposes and the development of AI applications.
The dataset comprises of descriptive metadata of 2.619.397 titles, which together form the "Alte Realkatalog" of Berlin State Libray, which may be translated to "Old Subject Catalogue". The data are stored in columnar format, containing 375 columns. They were downloaded in December 2023 from the German central library system (CBS). Exemplary tasks which can be served by this dataset comprise studies on the history of books between 1501 and 1955, on the paratextual formatting of scientific books between 1800 and 1955, and on pattern recognition on the basis of bibliographical metadata.
The primary intention for the publication of this dataset was the provision of a large computationally amenable dataset exclusively consisting of bibliographic metadata to stimulate research and development of AI applications. In 2024, large (meta-)datasets from the field of historical cultural data are still missing. In this respect, the dataset provided here aims to fill a gap. The dataset is suitable for the computational use of digitised and born-digital collections according to the Collections as Data principles.
The dataset was created by the team of the research project "Mensch.Maschine.Kultur – Künstliche Intelligenz für das Digitale Kulturelle Erbe" at Berlin State Library (SBB). The research project was funded by the Federal Government Commissioner for Culture and the Media (BKM), project grant no. 2522DIG002. The Minister of State for Culture and the Media is part of the German Federal Government.
Dataset Description
Repository
Zenodo Community: Data and Demos of the Staatsbibliothek zu Berlin – Berlin State Library
Website
Bibliographic Data from StaBiKat
Point of Contact
Sophie Schneider, Staatsbibliothek zu Berlin – Berlin State Library
Jörg Lehmann, Staatsbibliothek zu Berlin – Berlin State Library
Dataset Summary
Staatsbibliothek zu Berlin – Berlin State Library is an institution which belongs to the Prussian Cultural Heritage Foundation. It was founded in 1661 by the Elector of Brandenburg Friedrich Wilhelm and collects literature from all scientific fields, languages and countries. The systematic catalogue, known as the "Realkatalog", was established from 1842 onwards, based on a systematic listing of the contemporary holdings, new recordings of individual titles and the assignment of individual shelfmarks for each book. The result was a location-based systematic catalogue that accurately reflected the systematic arrangement of the books in the shelves. The final version of the systematic catalogue was completed in 1881. From then onwards, only new acquisitions were entered in the catalogue volumes. A reform of the arrangement and shelfmarks in 1946 meant that the catalogue was no longer bound to the arrangement in the magazines. However, it was continued as a systematic catalogue until 1955 for all printed matter (books, periodicals) published before 1956, including the several hundred thousand volumes that were added to replace the holdings that had not been returned from external storage during the Second World War. A new systematic catalogue has been started for the printed works published from 1956 onwards. The "Alte Realkatalog" (ARK), on the other hand, contains records of printed works that once belonged and still belong to the holdings of the Staatsbibliothek zu Berlin – Berlin State Library and its respective predecessors. Printed works published between 1501 and 1955 are included, and to this day, all antiquarian works from this period newly acquired by Berlin State Library are catalogued using the ARK system. Moreover, the digital record of titles which are digitised by libraries within the German federal system are added to the ARK in electronic form. As a side note, it is worth knowing that titles that once belonged to the library's collection but are classified as war losses are also contained in the dataset; the idea behind this approach is that the bibliographical record of the works is retained and the original collection context can still be recognised.
In a way that is peculiar for the time of its creation, the classification system of the ARK and the notations (i.e. signatures) correspond to each other. The system follows the traditional order of the faculties in the universities at the beginning of the 19th century. To identify the subjects and the books that belonged to them and were to be shelved with them, the subjects were subdivided using upper and lower case letters, with these letters corresponding to the shelfmarks. According to the classification begun in 1842, the seven main departments are subdivided as follows:
I. General (letter A, about 285.000 titles)
II. Theology (letters B to E, about 290.000 titles)
III. Political Science and Law (letters F to H, about 350.000 titles)
IV. Medicine, Natural Sciences (letters J to M, about 370.000 titles)
V. Pedagogy, Philosophy, Art, Technology (letters N to O, about 315.000 titles)
VI. History (letters P to U, about 590.000 titles)
VII. Language and Literature (letters V to Z, about 550.000 titles)
A detailed overview of the classification system and of the respective shelfmark groups can be found here online.
After duplicated entries were removed (some individual titles were listed in more than one letter of the classification system), the dataset comprises of about 2.6 million titles, the metadata of which are sorted into 375 columns.
The tasks for which this dataset may be used are studies on the history of books between 1501 and 1955, on the paratextual formatting of books between 1800 and 1955, and on pattern recognition on the basis of bibliographical metadata.
The dataset was created by systematically downloading all available titles of the ARK from the central library system (CBS). This was done letter by letter, according to the classification and notation system described above. The intention was to provide a substantial and at the same time delimited dataset comprising bibliographic metadata from Berlin State Library. The language used in the dataset is German; however, the titles described in the dataset may be in other languages as well, such as Latin, English, or French. According to the numbers of titles reported above for each of the major classes in the classification system, a broad range of domains, topics, and genres are covered by the titles of the "Alte Realkatalog". However, the emphasis of Berlin State Library on collecting literature relevant for the humanities is still identifiable.
Supported Tasks and Shared Tasks
There are three tasks for which this dataset may be used:
- Task "history of books between 1501 and 1955". See as an example Lathi et al, 2019
- Task "paratextual formatting of books between 1800 and 1955". See as an example Kilchör and Lehmann, 2022
- Task "pattern recognition on the basis of bibliographical metadata". Such patterns could be used for retrieval-augmented generation of metadata. See as an example Völker et al, 2024
These three tasks are narratively described in more detail in the file "Three Use Cases for the ARK Metadata" which is part of this dataset publication.
The dataset was not part of a shared task.
Languages
The languages represented in the dataset (i.e. the language of the publications listed there) are, in descending order of their frequency, German (ger), Latin (lat), French (fre), English (eng), Italian (ita), Russian (rus), Dutch (dut), Spanish (spa), Hebrew (heb), etc. Beyond these, several hundred other languages are represented as well. See the column "Sprachcodes" in the dataset, which specifies the language of the publication in ISO 639-2/B format.
Composition
The dataset does not contain any personal or sensitive information. It fully complies with the European General Data Protection Regulation (GDPR). The dataset does not contain information that is under copyright. It exclusively comprises of textual information and links. Authors and people who are subject of the listed titles are named and most often linked to authority files, especially to the Gemeinsame Normdatei (GND).
Dataset Structure
Within the German central library system (CBS), bibliographic information is provided in machine-readable form separated into fields. An overview of most of these fields as well as instructions on how information was inserted into each field can be found here (in German language only). Data for the 2.6 million titles were transformed from the format available in the CBS into a columnar format, where each of the fields forms an individual column. In the CBS, the field names are given in the form of four digits. For a better readability of the table, these four digits were replaced with the field names used in the documentation.
This dataset does not contain data splits which can be used for machine learning tasks. It is advisable to first analyse the dataset and then perform the split(s), for example to split off portions of the dataset relative to the numbers of titles published over time (see description of biases below).
Data Instances
For nearly every column, there can be found information in the documentation on how to interpret the content and resolve the keys used in every field. Field 1500, for example, holds the information of the language of the publication in ISO 639-2/B format, e.g. ger, lat, fre, eng. Moreover, further information has been inserted separated by a $ and a lower-case letter. The entry "ger$drus$deng$dfre" contains the separator $d and can be read as "the language of the publication is German, and there are summaries in Russian, English, and French". The entry "ger$cfre" can be read as "the language of the publication is German, but the original text is in French".
Data Fields
While most fields (or columns) are described in the documentation, there are a couple of columns in the dataset for which the documentation does not provide explanations. In the case of this dataset, these columns were populated by the Staatsbibliothek zu Berlin. The columns from "Schlagwortwortfolgen 5551" to "Schlagwortfolgen 5559" contain keywords which are specific to the "Alte Realkatalog." The columns from "Lokale Notationen 6000" to "Lokale Notationen 6089" contain notations from the "Alte Realkatalog" or the ARK classification system for the period 1501 to 1955; as a general rule, these are the basic signatures from the "Alte Realkatalog", followed by "ff.". The columns from "Lesesaalsystematik der SBB 6210" to "Lesesaalsystematik der SBB 6261" contain the classes used in the classification system of the "Alter Realkatalog" (see below). The columns "Lokale Schlagwörter 6500" to "Lokale Schlagwörter 6596" contain keywords provided by the Staatsbibliothek zu Berlin. The columns "Lokale Schlagwörter 6800" to "Lokale Schlagwörter 6808" contain information on the provenance of the titles, thus complementing the column "Provenienz". The column "Lokale Schlagwörter 6870" contains local standard data for coloured paper indexing. The columns "Lokale Schlagwörter 6880" to "Lokale Schlagwörter 6887" contain further information on previous owners and on acquisition, for example on books looted by the Nazis.
A key element in this dataset are Pica Production Numbers (PPNs). PPNs are unique identifiers used for individual entities; in the dataset, they are surrounded by exclamation marks (!...!) and are thus machine-readable. They are related to titles as well as to authorities. For example, the PPN 413526771 relates to the title "Das Feteh Mahari", a book about customs and laws of people in Eritrea. This PPN is used in the link via which the title can be accessed in the catalogue of the Berlin State Library (https://lbssbb.gbv.de/DB=1/XMLPRS=N/PPN?PPN=413526771). The title is part of a series, "Abhandlungen der Geistes- und Sozialwissenschaftlichen Klasse", which has received the PPN 129510084, so that all contributions to the series can be retrieved. The translator of this text, the Orientalist Prof. Dr. Maria Höfner, has received the PPN 134678230. It can be used to retrieve information about Maria Höfner via the online catalogue (https://lbssbb.gbv.de/DB=1/XMLPRS=N/PPN?PPN=134678230) or in the form of a xml file in DublinCore or MODS format. Her GND number is 118190415, her wikidata ID Q29016671. Furthermore, places, topics, and other authorities have also received PPNs, like the Tigre people (PPN 70000677X), an ethnic group indigenous to Eritrea, who are one of the subjects of the above-cited text. This PPN conforms to the GND number 4120412-8. Finally, all the classes in the classification system of the "Alter Realkatalog" have received PPNs as well. The PPN 086928120 is used for the sub-class "Literatur" of the subsection "Äthiopisch · Amharisch (Äthiopien) · Tigre (Äthiopien) · Ge'ez (Äthiopien)" which belongs to the class "Languages and Literature", and all the titles related to this sub-class can be retrieved via Berlin State Library's online catalogue as well using the PPN. In this entry, one can identify the notation range where this class is being used, in the chosen example signatures in the range between Zz 726 to Zz 845. This last type of PPNs is specific to the "Alte Realkatalog"; there is no GND equivalent for them, which documents that the "Old Subject Catalogue" is regarded as a classification system of its own.
Descriptive Statistics
The dataset comprises of 2.619.397 observations of 375 variables. In other words: 2.619.397 titles published between 1501 and 1955 are described in 375 columns. The columns are sparsely populated. Most of the columns are of the data type "character" due to additional, character-based codes within field contents, derived from the MARC-xml format which is stored in the central library system (CBS). The column "Pica Produktionsnummer", which holds the unique key for each title, is also in character format, because some of the keys contain an "X". File size is 960.784.112 Bytes.
Collection Process
The idea behind the decision to collect all the titles listed in the ARK in December 2023 was to compile a reasonably large dataset of bibliographic metadata which is at the same time delimitable and has not been gathered with regard to content. Rather, the criterion on whether or not a title is added to the ARK is a formal one, namely the year of publication, which has to lie between 1501 and 1955, or it is a republication or digital duplicate of a book that has been published in this period.
Alternative datasets with similar characteristics would be, for example, a compilation of a retrospective German national bibliography for historical prints of the 16th, 17th, and 18th century. Currently, the three national bibliographies VD 16, VD 17 and VD 18 (VD = "Verzeichnis der im deutschen Sprachraum erschienenen Drucke" of the respective centuries) comprise of 727.000 titles. A comprehensive publication of the metadata available for these titles might enable similar studies and tasks as have been described for the present dataset. A subset of the VD-dataset is contained in the present dataset; however, both datasets are not congruent.
Curation Rationale
The lack of massive datasets with bibliographic metadata motivated the creation of this dataset. Curation was performed with great care with the intent to keep as much information as possible while discarding some fields/columns due to them being sparsely populated and removing duplicated entries.
Source Data
Initial Data Collection and Normalisation
The data have been systematically downloaded from the central library system (CBS) in December 2023 and consolidated into one single table. During the process of consolidation, some columns were sorted out. In the vast majority of the cases, the decision for deletion was taken on a statistical basis: A column containing less than 10 entries seemed dispensable with regard to the fact that 2.620.000 entries could be there. Further reasons for the dismissal of columns were the fact that they are relevant only for internal purposes (e.g., the date of the last change of the bibliographic description) or where it was clear that they resulted from obvious mistakes.
After data collection and consolidation, data were modified only in three columns. In other words: In three columns, a part of the data were deleted. In these three cases, there were entries containing links to external sources (such as reviews or fulltexts of books) and it was specified that the transfer of the URL to third parties is not permitted. All other data are presented as is, without any manipulation.
No normalisation of any data was performed.
Who are the source data producers?
The source data were produced by trained librarians over the course of the past 180 years, i.e. starting from 1841. The titles to be included in the ARK were carefully selected over the course of several centuries, and the source data were created by librarians under changing rules and standards. It was only in 1899 that the so-called "Preußische Instruktionen" (Prussian instructions), a cataloguing set of rules for scientific libraries, were introduced. In the 20th century, they were replaced by other sets of rules, and in the second half of the 20th century the bibliographic records were transferred into electronic formats like MARC.xml and administred in the central library system CBS, where a broad range of sources and contributors were used to enrich the data. It is notable that this dataset does not only document a significant volume and diversity of publications, but is the result of centuries-long processes of selection of relevant historical titles, and of data collection conducted with care.
Data Provenance
The provenance of the data is the central library system (CBS), which contains the data of the union catalogue K10plus of the library networks GBV (Gemeinsamer Bibliotheksverbund) and SWB (Südwestdeutscher Bibliotheksverbund). Alternatively, the data could have been retrieved via the unAPI or SRU-API of the GBV. The licence generally attached to bibliographic metadata is CCO. However, this dataset has been curated by employees of the "Mensch.Maschine.Kultur" project and is published under a CC BY 4.0 license.
Use of Linked Open Data, Controlled Vocabulary, Multilingual Ontologies/Taxonomies
The dataset contains links to several authority file types (persons, places, subjects) and to classification systems such as the Nederlandse basisclassificatie (BK), the LCC or DDC notation systems. The classification was performed by the librarians who created the bibliographic record.
For example, people or families as creators of works are usually linked with the Gemeinsame Normdatei (GND, Integrated Autority File), like e.g. the author and contributor Karl Gustav Rodén, who has received the GND identifier 1300684216. The Basisklassifikation is usually linked via an PPN; e.g., the PPN 106404547 is used for the Basisklassifikation identifier 17.98 pointing to anthologies. The LCC notation is given in the original format, e.g. as HV8203 pointing to "Police, Detectives, Constabulary / By region or country". The same holds for the DDC Dewey Decimal Classification: The ID 360 points to the DDC class "Social Problems & Services".
Version Information
This is the first version of the dataset. The dataset has been collected in early December 2023. Even though the "Alte Realkatalog" has been completed following the publication year 1955, that does not mean that the number of titles contained in it stops to grow. Antiquarian works from the period 1501–1955 which are continually acquired by Berlin State Library are catalogued using the ARK system and are thus adding to the total number of titles which are contained in the dataset. Furthermore, the digital records of titles which have been digitised by other libraries within the network of German libraries and which have been published between 1500 and 1955 are being added to the ARK dataset. This also augments the dataset. However, it is currently not planned to publish an updated and enlarged version of all the titles contained in the "Alte Realkatalog".
Preprocessing and cleaning
Beyond the removal of duplicates and the data cleaning described above in the section "Initial Data Collection and Normalisation", no preprocessing of the dataset has been undertaken. The conversion of the dataset from data frame format to the columnar arrow format was performed with the aim to alleviate ingestion of the dataset for machine learning purposes.
Personal and Sensitive Information
The dataset does not contain personal or sensitive information beyond what is available in international authority files anyway. Authors of and contributors to the titles contained in this dataset may still be living; however, this complies with international privacy laws.
This dataset pertains to a difficult history insofar as it contains information on titles looted by the Nazis. This is a field of still ongoing research, and the database may serve as a source of further, especially quantitative, analysis.
Uses
The dataset is suitable to serve the three tasks described above. Further possible uses are provenance research, especially with respect to books and documents looted by the Nazis, or the creation of a knowledge graph out of the PPNs contained in the "Alte Realkatalog", and the related data which can be retrieved online as described above. Furthermore, the dataset can be used to examine the metadata diversity, the data quality, and its potential to being transformed into linked open data. People using this dataset are welcome to provide feedback on this dataset and of the uses made of it to the dataset curators.
Links to Related Datasets, Publications and Models
The dataset contains links to several other datasets which can be retrieved online as complimentary resources. See the section "Data fields" above, especially the explanations on the Pica Production Numbers (PPNs) used in the dataset.
Social Impact of Dataset
This dataset describes historical titles (printed books, monographs, multi-volume works, journals) that have been published between 1501 and 1955. The social impact of the dataset might therefore be very low. However, since a part of these publications refer to colonial subjects, i.e. the populations of the German colonies between 1879 and 1915, harmful effects on these populations are thinkable.
On a positive note, this dataset may be used to identify printed resources that contain valuable information on low-resource or underrepresented languages. These resources could be used to support underserved communities.
Discussion of Biases
This dataset definitely reflects at least two biases on different levels: The first bias is introduced by the classification system established in the second half of the 19th century – between 1842 and 1881 – and which is foundational for the "Alte Realkatalog". This system reflects the biases of these former times, especially with regard to biologistic thinking and dismissive consideration of specific population groups. Some examples of problematic categories in the "Alte Realkatalog" are as follows: The "physiology" subsection of the category "Medicine" lists "human races", and "negroes" and "Australians" as subcategories of "individual main human species". Consequently, there are also a subsections "eugenics" and "laws for the prevention of hereditary diseases" to be found in the category "Medicine". The "Africa" subsection of the category "History" lists "negroes, slave trade" as an own class, while the "economic science" subsection of the category "Political Science" lists "colonisation" as an own class. The "other occidental languages" subsection of the category "Language and Literature" lists "Gipsy language" as an own class, a classification which can be seen as a typical "other" category, and the "ethnography" subsection of the category "History" lists "population groups and races" as an own class. Taken together, these categories reveal the racist, discriminatory and colonialist thinking of the time viewing people from a range of 'scientific' perspectives (such as economy, medicine, or ethnography). Especially problematic is the fact that the biologistic foundation of the thinking implies a 'naturalisation' of the classification system which at the same time legitimises the power structures of the era of colonialism AND is the lingua franca of a library system.
The second bias to be named here is the uneven distribution of titles published over time. The biased classification system prepared in the second half of the 19th century was populated with the output of the rotary printing press, which was steeply increasing from the middle of the 19th century onwards and further accelerated with the invention and use of the offset press around 1870. While the absolute numbers of titles listed in the "Alte Realkatalog" stayed below 10.000 per year until about the 1870s, the numbers rise to about 20.000 per year around 1900, beyond 25.000 at the beginning of the First World War and beyond 35.000 per year at the beginning of the 1930s. While there has not yet been conducted a certainly illuminating analysis of how many titles were sorted into which category of the classification system over time, it seems most likely that, in terms of absolute numbers of titles published, this second bias exacerbates the first bias.
As the "Alte Realkatalog" has to be understood as a documentation of a historical evolution of printed matter, no steps were to taken to mitigate these biases. Rather, further analysis of the named biases and further ones are encouraged.
Other Known Limitations
There are no other known limitations of this dataset. Users are invited to report such limitations back to the curators.
Unanticipated Uses made of this Dataset
There are no known unanticipated uses made of this dataset. Users are invited to report the uses they made of this dataset back to the curators, which would enable an update of the datasheet.
Distribution
The dataset fully complies with the European General Data Protection Regulation GDPR.
Dataset Curators
The dataset was curated and published by two members of the research project "Mensch.Maschine.Kultur" ("Human.Machine.Culture"):
Sophie Schneider, Staatsbibliothek zu Berlin – Berlin State Library
Jörg Lehmann, Staatsbibliothek zu Berlin – Berlin State Library
Both curators can be contacted with regard to an update or feedback to the datasheet and regarding technical issues. The curators are prepared to incorporate responses and comments into a new version of the datasheet if this deems sensible.
Both curators are university graduates temporarily working in a research project located at Berlin State Library. Sophie Schneider has studied library and information science; she was responsible for the conversion of the data from plaintext into csv format. Jörg Lehmann has studied history and literature studies; he was responsible for downloading the data, curating them as described, and drafting the datasheet. The data contained in the dataset have been compiled by trained librarians over a long period of time and are therefore a typical result of carefully crafted metadata from a cultural heritage institution. The selection, transformation and curation of the dataset was performed with the goal to provide a computer-amenable dataset to both the research and machine learning communities.
Licensing Information
Creative Commons Attribution 4.0 International – CC BY 4.0
Citation Information
@dataset{lehmann_2024_DOI12783813,
author = {Schneider, Sophie and
Lehmann, Jörg},
title = {{Berlin State Library (2024). Metadata of the
"Alte Realkatalog (ARK)" of Berlin State Library
(SBB)}},
month = aug,
year = 2024,
publisher = {Zenodo},
version = {2024-08-09},
doi = {10.5281/zenodo.12783813},
url = {https://doi.org/10.5281/zenodo.12783813}
}
Berlin State Library (2024). Metadata of the "Alter Realkatalog (ARK)" of Berlin State Library (SBB). Zenodo. https://doi.org/10.5281/zenodo.12783813
Contributions
The "Alte Realkatalog" is a joint effort by librarians of the Staatsbibliothek zu Berlin – Berlin State Library since 1842 and all contributors to the central library system (CBS), through which the title data were retrieved in electronic form. However, as libraries have to reconsider their core tasks (the provision of metadata), this dataset publication transcends traditional librarian formats and provides a computer-amenable dataset.
Maintenance
The maintenance of this dataset is limited. The data will not be updated, but any technical issues will be addressed during the lifetime of the research project "Mensch.Maschine.Kultur", in the context of which this dataset was established. The project ends in June 2025, and the dataset will be maintained at least until then.
The dataset consists of a single file (ARK-Metadaten.parquet) as well as the accompanying datasheet. The .parquet format enables large amounts of data to be processed and moved quickly. It is a specific data format that stores data in a columnar memory layout and has been developed by Apache Arrow. Libraries are available for C, C++, C#, Go, Java, JavaScript, Julia, MATLAB, Python, R, Ruby, and Rust.
MD5 and SHA256 hashes of the file ARK-Metadaten.parquet:
MD5: e49380964d1df39c261e9d4e35f994b2
SHA256: eb01e50deb8da8499cc218512655e056e49df28e47903674874e91b2f7101b1a
Datasheet as of August 9th, 2024