metadata

configs:
  - config_name: DDC
    default: true
    data_files:
      - split: train
        path:
          - DDC.jsonl.gz
  - config_name: LCC
    data_files:
      - split: train
        path:
          - LCC.jsonl.gz
  - config_name: UDC
    data_files:
      - split: train
        path:
          - UDC.jsonl.gz
  - config_name: CLC
    data_files:
      - split: train
        path:
          - CLC.jsonl.gz
  - config_name: RVK
    data_files:
      - split: train
        path:
          - RVK.jsonl.gz
task_categories:
  - text-classification
task_ids:
  - multi-class-classification
pretty_name: Library Classification Systems
tags:
  - library-science
  - information-organization
  - ontology

Library Classification Systems

Dataset Summary

This comprehensive dataset encompasses hierarchical outlines of five major library classification systems:

Dewey Decimal Classification (DDC)
Library of Congress Classification (LCC)
Universal Decimal Classification (UDC)
Chinese Library Classification (CLC)
Regensburger Verbundklassifikation (RVK)

The dataset provides freely-available outlines of these classification systems, which are widely used in libraries and information management systems worldwide. It captures the hierarchical structure and descriptions for each classification, offering a valuable resource for researchers, librarians, and information scientists. However, it's important to note that the full text of the first three classification systems is not included and must be purchased separately from their organizations.

Data Format and Content

The data is structured in JSONL (JSON Lines) format, with each entry containing fields for:

Call numbers
Descriptions
Broader categories
Narrower subcategories

This format ensures easy parsing and integration into various data processing workflows.

Classification Systems Overview

DDC, LCC, and UDC: These are internationally recognized systems used in libraries across the globe.
CLC: Primarily used in China by public schools, libraries, and publishers.
RVK: Employed by several German universities for organizing their collections.

Supported Tasks and Leaderboards

This dataset is particularly valuable for a range of tasks in information science and machine learning, including:

Hierarchical Text Classification
Library Science and Information Organization
Ontology and Knowledge Graph Construction
Comparative Studies of Classification Systems
Machine Learning for Library Cataloging

While there are no specific leaderboards associated with this dataset, it presents opportunities for developing benchmarks in areas such as automated classification accuracy or ontology mapping.

Languages

The dataset incorporates multiple languages, reflecting the global nature of library classification systems:

English: DDC, LCC, and UDC
Simplified Chinese: CLC
German: RVK

This multilingual aspect enhances the dataset's utility for cross-lingual information retrieval and classification tasks.

Data Structure

Data Instances

Each instance in the dataset represents a library classification entry. The structure is designed to capture the hierarchical nature of the classification scheme.

Data Fields

call_number: string
- The alphanumeric call number that uniquely identifies the classification
- Example: "AC1-999"
description: string
- A textual description of the classification topic or subject area
- Example: "Collections. Series. Collected works"
broader: string or null
- The call number of the parent/broader classification
- Null for root-level classifications
- Example: null (for a top-level classification)
narrower: list of strings
- An array of call numbers representing child/narrower classifications
- Example: ["AC1-195", "AC200", "AC801-895", "AC901-995", "AC999"]

Example Instance

{
  "call_number": "AC1-999",
  "description": "Collections. Series. Collected works",
  "broader": null,
  "narrower": ["AC1-195", "AC200", "AC801-895", "AC901-995", "AC999"]
}

This structure allows for efficient navigation and querying of the classification hierarchy, enabling users to traverse from broader to narrower topics and vice versa.

Data Splits

The full dataset contains:

1,110 DDC entries
6,517 LCC entries
2,431 UDC entries
8,826 CLC entries
5,032 RVK entries

Total: 23,916 entries

Dataset Creation

Curation Rationale

This dataset was curated to provide open access to the hierarchical structure of major library classification systems. It aims to support research, education, and innovation in information organization and retrieval.

Source Data

The dataset is derived from the publicly available outlines of the Dewey Decimal Classification (DDC), Library of Congress Classification (LCC), and Universal Decimal Classification (UDC). Full classification texts are proprietary and require purchase from the respective organizations.

Initial Data Collection and Normalization

Classification outlines were collected on 2024-09-18 from the following authoritative sources:

Library of Congress Classification (LCC): Library of Congress Classification Outline
Universal Decimal Classification (UDC): Universal Decimal Classification Summary (ROSSIO Vocabulary Server)
Dewey Decimal Classification (DDC): Dewey Services - Resources (OCLC)
Chinese Library Classification (CLC): CLC Index
Regensburger Verbundklassifikation (RVK): RVK Online

Versions used:

UDC Master Reference File (UDC MRF) 2011
DDC 23 Summaries
RVK aktuelle csv-Datei (2023_2)

Data processing steps:

DDC: The summary PDF was downloaded and manually formatted for consistency.
LCC: Microsoft Word files were converted to plain text, with headings and call numbers extracted programmatically.
UDC: The udc-summary.rdf file was parsed using RDFlib, retaining only English labels.
CLC: The data was collected by crawling through the website. Only the first 3 levels of the hierarchy were included.
RVK: The CSV file was processed, keeping only the first 3 levels of the hierarchy.

Considerations for Using the Data

Social Impact of Dataset

This dataset expands access to library classification knowledge, potentially:

Facilitating research in information science and digital humanities
Supporting the development of improved information retrieval systems
Enhancing educational resources for library and information science students

Discussion of Biases

Users should be aware that library classification systems may reflect historical, cultural, and geographical biases in knowledge organization. These biases could manifest as:

Overrepresentation of Western perspectives
Outdated terminology or categorizations for certain topics
Uneven depth of coverage across different subject areas

Known Limitations

Scope: The dataset includes only freely available outlines, not the complete classification texts.
Granularity: The level of detail varies between classification systems and within subject areas.
Currency: Updates to the original classification systems may not be immediately reflected in this dataset.
Structural differences: The UDC system's structure, with its main and auxiliary tables, differs slightly from DDC and LCC, which may affect comparisons.

Datasets:

agentlans
/

library-classification-systems