metadata

license: odc-by
task_categories:
  - text-retrieval
language:
  - en
pretty_name: whatsthatbook
extra_gated_prompt: >-
  To access this dataset, you agree to the terms and conditions from the
  GoodReads website stated here: https://www.goodreads.com/about/terms
extra_gated_fields:
  I agree to use to the terms and conditions: checkbox

Dataset Card for WhatsThatBook

Dataset Description

Paper: https://arxiv.org/abs/2305.15053
Point of Contact: k-lin@berkeley.edu

Dataset Summary

A collection of tip-of-the-tongue queries for book searches. The dataset was curated from GoodReads community forum user queries. It seves as a training and evaluation resource for tip-of-the-tongue book queries. The user queries contain the interactions on the community forum and the documents are books with associated metadata.

Supported Tasks and Leaderboards

WhatsThatBook is intended for information retrieval tasks including but not limited to standard retrieval, using just the original query posted by the user and interactive settings, where the system asks clarification queries to narrow down the user's information needs.

Languages

The dataset is primary in English, some book descriptions may contain other languages.

Dataset Structure

Data Fields

Data fields for WhatsThatBook queries:

question: Inital query posted to the community forum
question_posted_date: The date that the query was posted in YYYY-MM-DD format
book_id: ID of the gold book used for evaluation
answers: List of the gold book descriptions

The fields for the books:

title: The title of the book
author: The author of the book
author_url: Link to the author page
description The blurb of the book that contains description of the plot or
isbn_13: The ISBN 13 number
date: String representation of the date from the book webpage
parsed_dates:A list of the publication date parsed out in YYYY-MM-DD format
image_link: original link to image
ratings: Total number of ratings
reviews: Total number of reviews
genres: Dictionary of genre tags to number of times tagged with that genre
id: ID of the book, corresponding to the query file

Data Splits

The dataset is comprised of two parts, WTB (WhatsThatBook), as well as TOMT (tip-of-my-tongue). WhatsThatBook contains standard train, dev, and test splits, and TOMT serves as an evaluation set.

Dataset Creation

Source Data

Additional Information

Dataset Curators

Kevin Lin, UC Berkeley, k-lin@berkeley.edu
Kyle Lo, Allen Institue For Artificial Intelligence, kylel@allenai.org

Citation Information

@article{lin2023decomposing,
  title={Decomposing Complex Queries for Tip-of-the-tongue Retrieval},
  author={Lin, Kevin and Lo, Kyle and Gonzalez, Joseph E and Klein, Dan},
  journal={arXiv preprint arXiv:2305.15053},
  year={2023}
}