pfaha/goodreads-books

Goodreads Books Metadata

Dataset Description

Goodreads Books Metadata is a structured dataset of book records scraped directly from Goodreads, a social platform for book readers and recommendations. The dataset is collected via an ongoing, resumable crawl and contains rich metadata per book: bibliographic information, crowd-sourced ratings, contributor (author/illustrator/editor/etc.) details enriched with author-level popularity stats, genre tags, series membership, and descriptive text.

The primary research objective behind this dataset is to understand how users rate books on Goodreads. This dataset is part of a personal data science portfolio project, demonstrating end-to-end data collection, transformation, and analysis on a real-world social platform.

• Homepage: https://www.goodreads.com
• Repository (source code): goodreads_books
• Point of Contact: pfaha (My Hugging Face profile)

---

Dataset Summary

Property	Value
Source	Goodreads Books (scraped)
Format	Parquet (partitioned, checkpointed every 5,000 books)
Language	Multilingual (book language field varies; primary UI/text is English)
Number of files	Growing incrementally, one books-partN.parquet file per checkpoint

This is a live, growing dataset: a queue-based crawler samples random book IDs, and every 5,000 successfully collected books are checkpointed to this repository as a new books-partN.parquet file. The target size is 500,000 valid book records; the dataset reflects however many checkpoints have completed so far.

---

Supported Tasks

This dataset can support various tasks, among which we identify the following:

• Book rating prediction — Predict a book's average_rating from other features in the dataset.
• Genre classification — Classify books into genre categories using description text and other metadata.
• Contributor influence analysis — Study how a contributor's num_followers and num_distinct_works correlate with average_rating, ratings_count, and reviews_count of their books.

---

Data Collection

Scraping pipeline

The dataset is collected using a custom asynchronous Python ETL pipeline built on Crawlee, and split into two stages:

1. EXTRACT (extract.py, class BookExtractor) — Drives one queue-based, self-healing sampling crawl per call:
   • Sampling: Random book IDs are drawn from a configured ID range, excluding any ID already tried in a prior run.
   • Fetching: Each sampled book ID's Goodreads page (https://www.goodreads.com/book/show/{id}) is fetched under a strict concurrency and rate-limit policy (see below).
   • Parsing: Each book page's __NEXT_DATA__ Apollo-cache JSON payload is parsed directly (not JSON-LD or raw CSS scraping) to extract structured book fields, contributors, genres, series membership, and identifiers.
   • Self-healing: Any terminal outcome for a book ID — successfully parsed, found to be nonexistent, flagged for a catalog guideline violation, or a permanently failed request — immediately triggers drawing and enqueuing one fresh replacement book ID, so the crawl queue never runs dry without a deliberate reason (ID space exhaustion).
   • Enrichment: For each book, the pipeline additionally fetches the first author's Goodreads profile page (for num_followers and num_distinct_works) and the work's editions-listing page (for num_editions and has_more_editions).
2. LOAD (load.py, class BookLoader) — Batches finished records into a Polars DataFrame, writes them locally as a Parquet file, uploads that file to this Hub repository, and updates a book_ids.parquet tracker file (also on the Hub) recording every book_id ever successfully scraped, so future runs never re-scrape the same book.

Politeness and rate limiting

The crawler enforces a hard cap of 60 requests per minute (1 request/second) against Goodreads, independent of local concurrency settings, specifically to respect Goodreads' request tolerance and avoid WAF-triggered blocking. Local concurrency (how many requests are in flight at once) is tuned separately and does not override this rate ceiling. Failed requests are retried automatically by the underlying crawler up to a fixed retry limit before being treated as a permanent failure and replaced with a fresh random book ID.

Data validity handling

A book ID is treated as nonexistent if its page yields no usable Apollo-cache book entry or no title. A book ID is separately flagged via catalog_guideline_violation if Goodreads has delisted it for a policy violation — text is present on the page but the book carries no usable bibliographic data; such books are still marked (not silently dropped) but are not treated as valid, populated records in the same way as normally parsed books.

---

Dataset Structure

Data Fields

The output strictly follows the schema produced by BookExtractor/BookLoader:

Column	Type	Description
book_id	string	Goodreads numeric book ID (kept as string).
url	string	Full Goodreads URL for the book page.
title	string	Book title as displayed on Goodreads.
contributors	list<struct>	Every contributor on the book (primary author first, then secondary contributors such as illustrators, translators, or editors). Each entry has name, url, role, num_followers, and num_distinct_works — the latter two are populated only for the first (primary) contributor, since only their author profile page is fetched for enrichment.
first_author	string	Convenience copy of the primary contributor's name (same as contributors[0].name).
description	string	Book description text, extracted from the book's Apollo-cache data and stripped of HTML markup.
genres	list<string>	List of genre tags as shown on the Goodreads book page.
series	list<struct>	Series membership, if any. Each entry has name (series title) and position (the book's position within that series, as a string since Goodreads positions can be non-integer, e.g. "1-4"). Empty list if the book is standalone.
format	string	Publication format (e.g. "Hardcover", "Paperback", "Ebook").
pages	int64	Page count for this edition.
language	string	Language of this edition, as reported by Goodreads.
asin	string	Amazon Standard Identification Number, if available.
isbn10	string	10-digit ISBN, if available.
isbn13	string	13-digit ISBN, if available.
publisher	string	Publisher name for this edition.
publication_date	string	ISO 8601 date string (YYYY-MM-DD) derived from the edition's publication timestamp.
average_rating	float64	Crowd-sourced average rating aggregated across all Goodreads user ratings for the work.
ratings_count	int64	Total number of ratings for the work.
reviews_count	int64	Total number of text reviews for the work.
work_id	int64	Goodreads internal work ID, used to derive num_editions/has_more_editions via the work's editions-listing page.
catalog_guideline_violation	bool	True if Goodreads has flagged this book ID as not meeting its catalog guidelines (delisted for policy reasons rather than never having existed).
num_editions	int64	Total number of editions of this work, taken from the work's editions-listing page.
has_more_editions	bool	True if num_editions is greater than 1.
is_part_of_series	bool	True if series is non-empty.

Data Splits

This dataset has no predefined train/validation/test split. All records are provided as a single collection partitioned across numbered Parquet files (books-part1.parquet, books-part2.parquet, etc.), each corresponding to one checkpoint of up to 5,000 newly collected books. Users should define their own splits depending on the downstream task.

---

Dataset Creation

Curation Rationale

Goodreads is a social reading platform hosting a large volume of user ratings, reviews, and contributors' profiles. This dataset was created to enable research into rating dynamics as expressed by a large and diverse random sample of the platform's catalog. It provides a real-world, richly structured corpus derived directly from Goodreads' own hydration data rather than fragile page-text scraping.

Source Data

All data originates from the public Goodreads web page. No authentication or account was required for access. Only publicly visible metadata was collected, and no user-identifiable information is present in this dataset.

Annotations

This dataset contains no manual annotations. All fields are extracted directly from Goodreads' own page data (the __NEXT_DATA__ Apollo cache, plus author and editions pages) and reflect the community-aggregated state of the platform at the time each book was scraped.

---

Considerations for Using this Dataset

Limitations

• Random-sampling coverage, not exhaustive coverage

  Books are discovered by drawing random IDs within a configured range, not by crawling Goodreads' catalog index.

  This means the dataset is a large random sample of the ID space, not a complete or guaranteed-representative census of Goodreads' full catalog.

• Snapshot in time, collected incrementally

  Different rows in this dataset were scraped at different times as the crawl progressed (checkpoints happen every 5,000 books).

  Fields like average_rating, ratings_count, reviews_count, and contributor num_followers/num_distinct_works change continuously on the live platform, so values reflect whatever point in time that specific book was scraped, not a single unified snapshot date.

• Contributor enrichment

  Primary contributor and secondary contributors' (illustrators, translators, co-authors, editors) Goodreads profiles are fetched for num_followers/num_distinct_works.

• No ratings distribution

  Individual star-rating breakdown counts (1★ to 5★) are not included — only the aggregated average_rating, ratings_count, and reviews_count.

• Catalog-guideline-violation books carry minimal data

  Rows where catalog_guideline_violation is true have no usable bibliographic fields populated (Goodreads has delisted the underlying content), and exist in the tracker/dataset primarily as a record that the ID was checked and excluded, not as a book with meaningful metadata.

• Social impact

  Findings derived from this data may not be representative of global reading preferences, since Goodreads' user base and catalog composition are not neutral samples of all books or all readers.

Licensing & Terms

This dataset was collected from publicly accessible Goodreads pages for research and educational purposes. Users of this dataset are responsible for complying with Goodreads' Terms of Service.

Do not use this dataset for commercial purposes without verifying applicable terms.

---

Citation

If you use this dataset in your research or projects, please cite it as:

@dataset{pfaha_goodreads_books_2026,
  author    = {pfaha},
  title     = {Goodreads Books},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/pfaha/goodreads-books}
}