PearlDiver / README.md
JoaoSchneider's picture
Update README.md
a70fcb9 verified
metadata
language:
  - de
license: cc-by-4.0
pretty_name: Perlentaucher
dataset_info:
  features:
    - name: date
      dtype: date32
    - name: author
      dtype: string
    - name: title
      dtype: string
    - name: ISBN
      dtype: string
    - name: price
      dtype: decimal128(6, 2)
    - name: n_pages
      dtype: uint16
    - name: n_reviews
      dtype: uint8
    - name: content
      dtype: string
    - name: publisher
      dtype: string
    - name: pub_place
      dtype: string
    - name: pub_year
      dtype: uint16
    - name: media_type
      dtype: string
    - name: media_spec
      dtype: string
    - name: is_relevant
      dtype: bool
    - name: is_novel
      dtype: bool
  splits:
    - name: train
      num_bytes: 33239971
      num_examples: 89790
  download_size: 15694656
  dataset_size: 33239971
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
tags:
  - Buchkritiken
  - Bücherschau des Tages
  - Books
  - Media
  - Book reviews

Listing of the >7k books I've read recently. Some claim it's actually data scraped from website Perlentaucher. Which you believe is up to you, but here's a video of my occasional reading activity (haters gonna say it's fake).


Dataset card

Dataset Description

The data set contains information of books that have been reviewed at least once. The reviews are received from reputable German print media such as the Frankfurter Allgemeine Zeitung (FAZ), Süddeutsche Zeitung (SZ), Die Zeit, etc. and other serious broadcasters such as Deutschlandfunk Kultur. Those reviews are collected by the culture magazine Perlentaucher and labeled as read a lot when there are at least three reviews of a book.

Perlentaucher publishes on a daily basis since the year 2000, except for German Sun- and holidays. On average, there are about M=12 (SD=5) bookentries per day. Data was harvested for all entries starting from March 15, 2000 to May 18, 2024. In total, the data set consists of 89,790 rows for 7,351 days and 15 columns.

Variables overview

  • date: the publication day of book review (dtype: date32, YYYY-MM-DD)
  • author: the author of the book (dtype: string)
  • title: the title of the book (dtype: string)
  • ISBN: the International Standard Book Number, ISBN-13 in normalized form (i.e. hyphens removed); correctness has been tested via check digit; missing/incorrect numbers have been corrected manually (dtype: string)
  • price: the price of the book in euros (decimal128(6, 2), i.e. scale = 2 for cents)
  • n_pages: the number of book pages; nan for other types such as audio files (dtype: uint16)
  • n_reviews: the number of review notes as an indicator whether the book is relevant; lumped to 1 for n<3 notes (dtype: uint8)
  • content: the dust cover blurb (dtype: string)
  • publisher: the publisher of the book (dtype: string)
  • pub_place: the place (city) where the books have been published; incorrect/inconsistent spellings have been corrected manually (dtype: string)
  • pub_year: the year the books has been published (dtype: uint16)
  • media_type: the type of media, i.e. audio, print, videofilm or unknown (dtype: string)
  • media_spec: a specification of the media type; for books its binding like soft- or hardcover (dtype: string)
  • is_relevant: boolean, whether the book is relevant (i.e. marked as read a lot, dtype: bool)
  • is_novel: boolean, whether the book is a novel according to its title (dtype: bool)

Citation Information


@misc {schneider2024pearldive,
    author       = { Schneider, Jo\~{a}o A G },
    title        = { Pearl Diver },
    year         = { 2024 },
    month        = { May },
    url          = { https://huggingface.co/datasets/JoaoSchneider/PearlDiver },
    doi          = { 10.57967/hf/2245 },
    publisher    = { Hugging Face },
    note         = { Revision 786df49 }
}