TheBritishLibrary/blbooksgenre

Name: TheBritishLibrary/blbooksgenre
Creator: TheBritishLibrary
Published: 2026-04-07 14:54:34
License: 暂无描述

Hugging Face2026-04-07 更新2024-03-04 收录

下载链接：

https://hf-mirror.com/datasets/TheBritishLibrary/blbooksgenre

下载链接

链接失效反馈

官方服务：

资源简介：

--- annotations_creators: - expert-generated language_creators: - crowdsourced - expert-generated language: - de - en - fr - nl license: - cc0-1.0 multilinguality: - multilingual size_categories: - 10K<n<100K - 1K<n<10K source_datasets: - original task_categories: - text-classification - text-generation - fill-mask task_ids: - topic-classification - multi-label-classification - language-modeling - masked-language-modeling pretty_name: British Library Books Genre dataset_info: - config_name: title_genre_classifiction features: - name: BL record ID dtype: string - name: title dtype: string - name: label dtype: class_label: names: '0': Fiction '1': Non-fiction splits: - name: train num_bytes: 187600 num_examples: 1736 download_size: 20111420 dataset_size: 187600 - config_name: annotated_raw features: - name: BL record ID dtype: string - name: Name dtype: string - name: Dates associated with name dtype: string - name: Type of name dtype: string - name: Role dtype: string - name: All names sequence: string - name: Title dtype: string - name: Variant titles dtype: string - name: Series title dtype: string - name: Number within series dtype: string - name: Country of publication sequence: string - name: Place of publication sequence: string - name: Publisher dtype: string - name: Date of publication dtype: string - name: Edition dtype: string - name: Physical description dtype: string - name: Dewey classification dtype: string - name: BL shelfmark dtype: string - name: Topics dtype: string - name: Genre dtype: string - name: Languages sequence: string - name: Notes dtype: string - name: BL record ID for physical resource dtype: string - name: classification_id dtype: string - name: user_id dtype: string - name: subject_ids dtype: string - name: annotator_date_pub dtype: string - name: annotator_normalised_date_pub dtype: string - name: annotator_edition_statement dtype: string - name: annotator_FAST_genre_terms dtype: string - name: annotator_FAST_subject_terms dtype: string - name: annotator_comments dtype: string - name: annotator_main_language dtype: string - name: annotator_other_languages_summaries dtype: string - name: annotator_summaries_language dtype: string - name: annotator_translation dtype: string - name: annotator_original_language dtype: string - name: annotator_publisher dtype: string - name: annotator_place_pub dtype: string - name: annotator_country dtype: string - name: annotator_title dtype: string - name: Link to digitised book dtype: string - name: annotated dtype: bool - name: Type of resource dtype: class_label: names: '0': Monograph '1': Serial - name: created_at dtype: timestamp[s] - name: annotator_genre dtype: class_label: names: '0': Fiction '1': Can't tell '2': Non-fiction '3': The book contains both Fiction and Non-Fiction splits: - name: train num_bytes: 3583138 num_examples: 4398 download_size: 20111420 dataset_size: 3583138 - config_name: raw features: - name: BL record ID dtype: string - name: Name dtype: string - name: Dates associated with name dtype: string - name: Type of name dtype: string - name: Role dtype: string - name: All names sequence: string - name: Title dtype: string - name: Variant titles dtype: string - name: Series title dtype: string - name: Number within series dtype: string - name: Country of publication sequence: string - name: Place of publication sequence: string - name: Publisher dtype: string - name: Date of publication dtype: string - name: Edition dtype: string - name: Physical description dtype: string - name: Dewey classification dtype: string - name: BL shelfmark dtype: string - name: Topics dtype: string - name: Genre dtype: string - name: Languages sequence: string - name: Notes dtype: string - name: BL record ID for physical resource dtype: string - name: classification_id dtype: string - name: user_id dtype: string - name: subject_ids dtype: string - name: annotator_date_pub dtype: string - name: annotator_normalised_date_pub dtype: string - name: annotator_edition_statement dtype: string - name: annotator_FAST_genre_terms dtype: string - name: annotator_FAST_subject_terms dtype: string - name: annotator_comments dtype: string - name: annotator_main_language dtype: string - name: annotator_other_languages_summaries dtype: string - name: annotator_summaries_language dtype: string - name: annotator_translation dtype: string - name: annotator_original_language dtype: string - name: annotator_publisher dtype: string - name: annotator_place_pub dtype: string - name: annotator_country dtype: string - name: annotator_title dtype: string - name: Link to digitised book dtype: string - name: annotated dtype: bool - name: Type of resource dtype: class_label: names: '0': Monograph '1': Serial '2': Monographic component part - name: created_at dtype: string - name: annotator_genre dtype: string splits: - name: train num_bytes: 27518816 num_examples: 55343 download_size: 20111420 dataset_size: 27518816 config_names: - annotated_raw - raw - title_genre_classifiction configs: - config_name: title_genre_classifiction data_files: - split: train path: data/title_genre_classifiction/*.parquet - config_name: annotated_raw data_files: - split: train path: data/annotated_raw/*.parquet - config_name: raw data_files: - split: train path: data/raw/*.parquet default_config_name: title_genre_classifiction --- # Dataset Card for blbooksgenre ## Table of Contents - [Dataset Card for blbooksgenre](#dataset-card-for-blbooksgenre) - [Table of Contents](#table-of-contents) - [Dataset Description](#dataset-description) - [Dataset Summary](#dataset-summary) - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards) - [Supervised tasks](#supervised-tasks) - [Languages](#languages) - [Dataset Structure](#dataset-structure) - [Data Instances](#data-instances) - [Data Fields](#data-fields) - [Data Splits](#data-splits) - [Dataset Creation](#dataset-creation) - [Curation Rationale](#curation-rationale) - [Source Data](#source-data) - [Initial Data Collection and Normalization](#initial-data-collection-and-normalization) - [Who are the source language producers?](#who-are-the-source-language-producers) - [Annotations](#annotations) - [Annotation process](#annotation-process) - [Who are the annotators?](#who-are-the-annotators) - [Personal and Sensitive Information](#personal-and-sensitive-information) - [Considerations for Using the Data](#considerations-for-using-the-data) - [Social Impact of Dataset](#social-impact-of-dataset) - [Discussion of Biases](#discussion-of-biases) - [Colonialism](#colonialism) - [Other Known Limitations](#other-known-limitations) - [Additional Information](#additional-information) - [Dataset Curators](#dataset-curators) - [Licensing Information](#licensing-information) - [Citation Information](#citation-information) - [Contributions](#contributions) ## Dataset Description - **Homepage:**: [https://doi.org/10.23636/BKHQ-0312](https://doi.org/10.23636/BKHQ-0312) - **Repository:** [https://doi.org/10.23636/BKHQ-0312](https://doi.org/10.23636/BKHQ-0312) - **Paper:** - **Leaderboard:** - **Point of Contact:** ### Dataset Summary This dataset consists of metadata relating to books [digitised by the British Library in partnership with Microsoft](https://www.bl.uk/collection-guides/google-books-digitised-printed-heritage). Some of this metadata was exported from the British Library catalogue whilst others was generated as part of a crowdsourcing project. The text of this book and other metadata can be found on the [date.bl](https://data.bl.uk/bl_labs_datasets/#3) website. The majority of the books in this collection were published in the 18th and 19th Century but the collection also includes a smaller number of books from earlier periods. Items within this collection cover a wide range of subject areas including geography, philosophy, history, poetry and literature and are published in a variety of languages. For the subsection of the data which contains additional crowsourced annotations the date of publication breakdown is as follows: | | Date of publication | | ---- | ------------------- | | 1630 | 8 | | 1690 | 4 | | 1760 | 10 | | 1770 | 5 | | 1780 | 5 | | 1790 | 18 | | 1800 | 45 | | 1810 | 96 | | 1820 | 152 | | 1830 | 182 | | 1840 | 259 | | 1850 | 400 | | 1860 | 377 | | 1870 | 548 | | 1880 | 776 | | 1890 | 1484 | | 1900 | 17 | | 1910 | 1 | | 1970 | 1 | [More Information Needed] ### Supported Tasks and Leaderboards The digitised books collection which this dataset describes has been used in a variety of digital history and humanities projects since being published. This dataset is suitable for a variety of unsupervised tasks and for a 'genre classification task'. #### Supervised tasks The main possible use case for this dataset is to develop and evaluate 'genre classification' models. The dataset includes human generated labels for whether a book is 'fiction' or 'non-fiction'. This has been used to train models for genre classifcation which predict whether a book is 'fiction' or 'non-fiction' based on its title. ### Languages [More Information Needed] ## Dataset Structure The dataset currently has three configurations intended to support a range of tasks for which this dataset could be used for: - `title_genre_classifiction` : this creates a de-duplicated version of the dataset with the `BL record`, `title` and `label`. - `annotated_raw`: This version of the dataset includes all fields from the original dataset which are annotated. This includes duplication from different annotators" - `raw`: This version of the dataset includes all the data from the original data including data without annotations. ### Data Instances An example data instance from the `title_genre_classifiction` config: ```python {'BL record ID': '014603046', 'title': 'The Canadian farmer. A missionary incident [Signed: W. J. H. Y, i.e. William J. H. Yates.]', 'label': 0} ``` An example data instance from the `annotated_raw` config: ```python {'BL record ID': '014603046', 'Name': 'Yates, William Joseph H.', 'Dates associated with name': '', 'Type of name': 'person', 'Role': '', 'All names': ['Yates, William Joseph H. [person] ', ' Y, W. J. H. [person]'], 'Title': 'The Canadian farmer. A missionary incident [Signed: W. J. H. Y, i.e. William J. H. Yates.]', 'Variant titles': '', 'Series title': '', 'Number within series': '', 'Country of publication': ['England'], 'Place of publication': ['London'], 'Publisher': '', 'Date of publication': '1879', 'Edition': '', 'Physical description': 'pages not numbered, 21 cm', 'Dewey classification': '', 'BL shelfmark': 'Digital Store 11601.f.36. (1.)', 'Topics': '', 'Genre': '', 'Languages': ['English'], 'Notes': 'In verse', 'BL record ID for physical resource': '004079262', 'classification_id': '267476823.0', 'user_id': '15.0', 'subject_ids': '44369003.0', 'annotator_date_pub': '1879', 'annotator_normalised_date_pub': '1879', 'annotator_edition_statement': 'NONE', 'annotator_FAST_genre_terms': '655 7 ‡aPoetry‡2fast‡0(OCoLC)fst01423828', 'annotator_FAST_subject_terms': '60007 ‡aAlice,‡cGrand Duchess, consort of Ludwig IV, Grand Duke of Hesse-Darmstadt,‡d1843-1878‡2fast‡0(OCoLC)fst00093827', 'annotator_comments': '', 'annotator_main_language': '', 'annotator_other_languages_summaries': 'No', 'annotator_summaries_language': '', 'annotator_translation': 'No', 'annotator_original_language': '', 'annotator_publisher': 'NONE', 'annotator_place_pub': 'London', 'annotator_country': 'enk', 'annotator_title': 'The Canadian farmer. A missionary incident [Signed: W. J. H. Y, i.e. William J. H. Yates.]', 'Link to digitised book': 'http://access.bl.uk/item/viewer/ark:/81055/vdc_00000002842E', 'annotated': True, 'Type of resource': 0, 'created_at': datetime.datetime(2020, 8, 11, 14, 30, 33), 'annotator_genre': 0} ``` ### Data Fields The data fields differ slightly between configs. All possible fields for the `annotated_raw` config are listed below. For the `raw` version of the dataset datatypes are usually string to avoid errors when processing missing values. - `BL record ID`: an internal ID used by the British Library, this can be useful for linking this data to other BL collections. - `Name`: name associated with the item (usually author) - `Dates associated with name`: dates associated with above e.g. DOB - `Type of name`: whether `Name` is a person or an organization etc. - `Role`: i.e. whether `Name` is `author`, `publisher` etc. - `All names`: a fuller list of names associated with the item. - `Title`: The title of the work - `Variant titles` - `Series title` - `Number within series` - `Country of publication`: encoded as a list of countries listed in the metadata - `Place of publication`: encoded as a list of places listed in the metadata - `Publisher` - `Date of publication`: this is encoded as a string since this field can include data ranges i.e.`1850-1855`. - `Edition` - `Physical description`: encoded as a string since the format of this field varies - `Dewey classification` - `BL shelfmark`: a British Library shelf mark - `Topics`: topics included in the catalogue record - `Genre` the genre information included in the original catalogue record note that this is often missing - `Languages`; encoded as a list of languages - `Notes`: notes from the catalogue record - `BL record ID for physical resource` The following fields are all generated via the crowdsourcing task (discussed in more detail below) - `classification_id`: ID for the classification in the annotation task - `user_id` ID for the annotator - `subject_ids`: internal annotation task ID - `annotator_date_pub`: an updated publication data - `annotator_normalised_date_pub`: normalized version of the above - `annotator_edition_statement` updated edition - `annotator_FAST_genre_terms`: [FAST classification genre terms](https://www.oclc.org/research/areas/data-science/fast.html) - `annotator_FAST_subject_terms`: [FAST subject terms](https://www.oclc.org/research/areas/data-science/fast.html) - `annotator_comments`: free form comments - `annotator_main_language` - `annotator_other_languages_summaries` - `'annotator_summaries_language` - `annotator_translation` - `annotator_original_language` - `annotator_publisher` - `annotator_place_pub` - `annotator_country` - `annotator_title` - `Link to digitised book` - `annotated`: `bool` flag to indicate if row has annotations or not - `created_at`: when the annotation was created - `annotator_genre`: the updated annotation for the `genre` of the book. Finally the `label` field of the `title_genre_classifiction` configuration is a class label with values 0 (Fiction) or 1 (Non-fiction). [More Information Needed] ### Data Splits This dataset contains a single split `train`. ## Dataset Creation **Note** this section is a work in progress. ### Curation Rationale The books in this collection were digitised as part of a project partnership between the British Library and Microsoft. [Mass digitisation](https://en.wikipedia.org/wiki/Category:Mass_digitization) i.e. projects where there is a goal to quickly digitise large volumes of materials shape the selection of materials to include in a number of ways. Some consideratoins which are often involved in the decision of whether to include items for digitization include (but are not limited to): - copyright status - preservation needs- the size of an item, very large and very small items are often hard to digitize quickly These criteria can have knock-on effects on the makeup of a collection. For example systematically excluding large books may result in some types of book content not being digitized. Large volumes are likely to be correlated to content to at least some extent so excluding them from digitization will mean that material is under represented. Similarly copyright status is often (but not only) determined by publication data. This can often lead to a rapid fall in the number of items in a collection after a certain cut-off date. All of the above is largely to make clear that this collection was not curated with the aim of creating a representative sample of the British Library's holdings. Some material will be over-represented and other under-represented. Similarly, the collection should not be considered a representative sample of what was published across the time period covered by the dataset (nor that that the relative proportions of the data for each time period represent a proportional sample of publications from that period). [More Information Needed] ### Source Data The original source data (physical items) includes a variety of resources (predominantly monographs) held by the [British Library](bl.uk/](https://bl.uk/). The British Library is a [Legal Deposit](https://www.bl.uk/legal-deposit/about-legal-deposit) library. "Legal deposit requires publishers to provide a copy of every work they publish in the UK to the British Library. It's existed in English law since 1662."[source](https://www.bl.uk/legal-deposit/about-legal-deposit). [More Information Needed] #### Initial Data Collection and Normalization This version of the dataset was created partially from data exported from British Library catalogue records and partially via data generated from a crowdsourcing task involving British Library staff. #### Who are the source language producers? [More Information Needed] ### Annotations The data does includes metadata associated with the books these are produced by British Library staff. The additional annotations were carried out during 2020 as part of an internal crowdsourcing task. #### Annotation process New annotations were produced via a crowdsourcing tasks. Annotators have the option to pick titles from a particular language subset from the broader digitized 19th century books collection. As a result the annotations are not random and overrepresent some languages. [More Information Needed] #### Who are the annotators? Staff working at the British Library. Most of these staff work with metadata as part of their jobs and so could be considered expert annotators. [More Information Needed] ### Personal and Sensitive Information [More Information Needed] ## Considerations for Using the Data There a range of considerations around using the data. These include the representativeness of the dataset, the bias towards particular languages etc. It is also important to note that library metadata is not static. The metadata held in library catalogues is updated and changed over time for a variety of reasons. The way in which different institutions catalogue items also varies. As a result it is important to evaluate the performance of any models trained on this data before applying to a new collection. [More Information Needed] ### Social Impact of Dataset [More Information Needed] ### Discussion of Biases The text in this collection is derived from historic text. As a result the text will reflect to social beliefs and attitudes of this time period. The titles of the book give some sense of their content. Examples of book titles which appear in the data (these are randomly sampled from all titles): - 'Rhymes and Dreams, Legends of Pendle Forest, and other poems', - "Précis of Information concerning the Zulu Country, with a map. Prepared in the Intelligence Branch of the Quarter-Master-General's Department, Horse Guards, War Office, etc", - 'The fan. A poem', - 'Grif; a story of Australian Life', - 'Calypso; a masque: in three acts, etc', - 'Tales Uncle told [With illustrative woodcuts.]', - 'Questings', - 'Home Life on an Ostrich Farm. With ... illustrations', - 'Bulgarya i Bulgarowie', - 'Εἰς τα βαθη της Ἀφρικης [In darkest Africa.] ... Μεταφρασις Γεωρ. Σ. Βουτσινα, etc', - 'The Corsair, a tale', 'Poems ... With notes [With a portrait.]', - 'Report of the Librarian for the year 1898 (1899, 1901, 1909)', - "The World of Thought. A novel. By the author of 'Before I began to speak.'", - 'Amleto; tragedia ... recata in versi italiani da M. Leoni, etc'] Whilst using titles alone, is obviously insufficient to integrate bias in this collection it gives some insight into the topics covered by books in the corpus. Further looking into the tiles highlight some particular types of bias we might find in the collection. This should in no way be considered an exhaustive list. #### Colonialism We can see even in the above random sample of titles examples of colonial attitudes. We can try and interrogate this further by searching for the name of countries which were part of the British Empire at the time many of these books were published. Searching for the string `India` in the titles and randomly sampling 10 titles returns: - "Travels in India in the Seventeenth Century: by Sir Thomas Roe and Dr. John Fryer. Reprinted from the 'Calcutta Weekly Englishman.'", - 'A Winter in India and Malaysia among the Methodist Missions', - "The Tourist's Guide to all the principal stations on the railways of Northern India [By W. W.] ... Fifth edition", - 'Records of Sport and Military Life in Western India ... With an introduction by ... G. B. Malleson', - "Lakhmi, the Rájpút's Bride. A tale of Gujarát in Western India [A poem.]", - 'The West India Commonplace Book: compiled from parliamentary and official documents; shewing the interest of Great Britain in its Sugar Colonies', - "From Tonkin to India : by the sources of the Irawadi, January '95-January '96", - 'Case of the Ameers of Sinde : speeches of Mr. John Sullivan, and Captain William Eastwick, at a special court held at the India House, ... 26th January, 1844', - 'The Andaman Islands; their colonization, etc. A correspondence addressed to the India Office', - 'Ancient India as described by Ptolemy; being a translation of the chapters which describe India and Eastern Asia in the treatise on Geography written by Klaudios Ptolemaios ... with introduction, commentary, map of India according to Ptolemy, and ... index, by J. W. McCrindle'] Searching form the string `Africa` in the titles and randomly sampling 10 titles returns: - ['De Benguella ás Terras de Iácca. Descripção de uma viagem na Africa Central e Occidental ... Expedição organisada nos annos de 1877-1880. Edição illustrada', - 'To the New Geographical Society of Edinburgh [An address on Africa by H. M. Stanley.]', - 'Diamonds and Gold in South Africa ... With maps, etc', - 'Missionary Travels and Researches in South Africa ... With notes by F. S. Arnot. With map and illustrations. New edition', - 'A Narrative of a Visit to the Mauritius and South Africa ... Illustrated by two maps, sixteen etchings and twenty-eight wood-cuts', - 'Side Lights on South Africa ... With a map, etc', - 'My Second Journey through Equatorial Africa ... in ... 1886 and 1887 ... Translated ... by M. J. A. Bergmann. With a map ... and ... illustrations, etc', - 'Missionary Travels and Researches in South Africa ... With portrait and fullpage illustrations', - '[African sketches.] Narrative of a residence in South Africa ... A new edition. To which is prefixed a biographical sketch of the author by J. Conder', - 'Lake Ngami; or, Explorations and discoveries during four years wandering in the wilds of South Western Africa ... With a map, and numerous illustrations, etc'] Whilst this dataset doesn't include the underlying text it is important to consider the potential attitudes represented in the title of the books, or the full text if you are using this dataset in conjunction with the full text. [More Information Needed] ### Other Known Limitations [More Information Needed] ## Additional Information ### Dataset Curators [More Information Needed] ### Licensing Information The books are licensed under the [CC Public Domain Mark 1.0](https://creativecommons.org/publicdomain/mark/1.0/) license. ### Citation Information ```bibtex @misc{british library_genre, title={ 19th Century Books - metadata with additional crowdsourced annotations}, url={https://doi.org/10.23636/BKHQ-0312}, author={{British Library} and Morris, Victoria and van Strien, Daniel and Tolfo, Giorgia and Afric, Lora and Robertson, Stewart and Tiney, Patricia and Dogterom, Annelies and Wollner, Ildi}, year={2021}} ``` ### Contributions Thanks to [@davanstrien](https://github.com/davanstrien) for adding this dataset.

提供机构：

TheBritishLibrary

原始信息汇总

数据集概述

数据集名称

名称: British Library Books Genre
别名: blbooksgenre

数据集特点

语言: 包含德语（de）、英语（en）、法语（fr）、荷兰语（nl）
许可证: cc0-1.0
多语言性: 多语言
大小: 数据集大小分为1K<n<10K和10K<n<100K两个类别
源数据集: 原始数据
任务类别: 文本分类、文本生成、填空
任务ID: 主题分类、多标签分类、语言建模、掩码语言建模

数据集结构

配置名称:
- title_genre_classifiction
- annotated_raw
- raw
特征:
- title_genre_classifiction: 包含BL记录ID、标题、标签（分类为小说或非小说）
- annotated_raw: 包含详细的书籍元数据，如出版信息、作者信息、分类信息等
- raw: 包含原始的书籍元数据，无额外注释

数据实例

title_genre_classifiction: 示例包含BL记录ID、标题和标签
annotated_raw: 示例包含详细的书籍元数据和注释信息

数据字段

通用字段: BL记录ID、标题、出版日期、出版地、语言等
注释字段: 包括注释者提供的出版日期、分类、主题等更新信息

数据分割

分割名称: train
大小:
- title_genre_classifiction: 1736个实例，187600字节
- annotated_raw: 4398个实例，3583138字节
- raw: 55343个实例，27518816字节

数据创建

注释创建者: 专家生成、众包
语言创建者: 众包、专家生成
注释过程: 通过内部众包任务完成，注释者为英国图书馆工作人员

使用数据集的考虑因素

数据集代表性: 数据集可能不具有完全的代表性，某些类型的书籍内容可能未被充分数字化
语言偏见: 注释数据可能过度代表某些语言
元数据动态性: 图书馆元数据会随时间更新和变化，使用前需评估模型的泛化能力

搜集汇总

数据集介绍

构建方式

该数据集源自大英图书馆与微软合作开展的图书数字化项目，整合了图书馆目录导出的元数据与内部众包标注的补充信息。构建过程中，首先从数字化馆藏中提取了18至19世纪为主的图书记录，涵盖地理、哲学、历史、诗歌及文学等多领域，涉及英语、德语、法语、荷兰语等多种语言。随后，通过面向图书馆专业人员的众包任务，对部分记录进行了精细化标注，包括对体裁（小说或非小说）、出版信息、主题术语等字段的修正与丰富。最终，数据集被组织为三个配置：精简的标题体裁分类版、包含完整标注信息的详细版以及未经筛选的原始版，以适应不同研究需求。

特点

该数据集的核心特色在于其多层次、多维度的元数据结构。它提供了从基础书目信息（如标题、作者、出版地）到深度标注内容（如FAST体裁与主题术语、注释者评论）的丰富字段，尤其以体裁分类标签（小说/非小说）为亮点，支持监督式学习任务。数据集的众包标注由具备元数据专业知识的图书馆员工完成，确保了标注质量。此外，其多语言覆盖与跨世纪的时间跨度，为数字人文研究中的历史文本分析提供了独特资源，但需注意数据并非随机样本，存在语言和出版年代上的选择偏差。

使用方法

该数据集可通过HuggingFace Datasets库便捷加载，支持三种配置的选择：'title_genre_classifiction'适用于基于标题的体裁分类任务，'annotated_raw'适合需要完整元数据及众包标注的深度分析，而'raw'则保留了所有原始记录以供探索。用户可指定配置名称（如dataset = load_dataset('TheBritishLibrary/blbooksgenre', 'title_genre_classifiction')）获取对应子集。数据以Parquet格式存储，仅提供训练集划分，可直接用于文本分类、语言建模或掩码语言建模等任务。建议在使用前注意数据偏差，如标注非随机性及年代分布不均，并参考CC0-1.0许可协议进行引用。

背景与挑战

背景概述

在数字人文与计算语言学交叉领域，大规模历史文献元数据的结构化标注为文本分类研究提供了独特契机。British Library Books Genre数据集由大英图书馆于2020年创建，依托其与微软合作开展的馆藏数字化项目，旨在通过众包标注与专家审核相结合的方式，解决18至19世纪多语种图书的体裁自动识别问题。该数据集收录了超过五万条书目记录，涵盖英、德、法、荷等语言，核心研究问题聚焦于如何从标题等有限元数据中区分虚构与非虚构作品。其影响力体现在为数字历史研究提供了可复用的基准资源，并推动了面向历史文本的弱监督分类方法发展。

当前挑战

该数据集面临的核心挑战在于领域问题的复杂性与构建过程的偏差。首先，历史文本体裁分类需应对元数据稀疏性问题——原始目录记录中体裁字段常缺失，且18至19世纪出版物的标题风格与现代差异显著，增加了基于标题的分类难度。其次，众包标注过程中存在语言子集非随机选取的偏差，导致德语、法语等语种样本过度集中，而早期出版物（如17世纪）数量稀少。此外，数字化筛选标准（如排除超大型图书）造成馆藏代表性不足，使模型可能学习到与内容无关的出版年代或物理形态特征，削弱跨时段泛化能力。

常用场景

经典使用场景

在数字人文与计算语言学领域，British Library Books Genre数据集为基于标题的图书体裁二元分类（虚构与非虚构）提供了经典基准。该数据集包含逾五千余条经由图书馆专家标注的元数据记录，覆盖18至19世纪多语种出版物，其简洁的标题-标签结构使得研究者能够高效训练并评估文本分类模型，尤其适用于低资源场景下的迁移学习与跨语言分类任务。

衍生相关工作

该数据集催生了多项开创性工作，包括基于Transformer架构的标题体裁预测模型（如GenreBERT），以及融合多模态元数据（如出版地、作者信息）的体裁分类框架。部分研究还将其与FAST主题词表结合，探索体裁与主题的联合建模，为数字人文中的知识图谱构建与文献计量分析提供了新范式。

数据集最近研究