Dataset Card for QUEST

Dataset Summary

We provide here the data accompanying the paper: QUEST: A Retrieval Dataset of Entity-Seeking Queries with Implicit Set Operations .

Dataset Structure

Data Instances

QUEST contains 6307 training queries, 323 examples for development, and 1727 examples for testing.

Data Fields

Each examples file contains newline-separated json dictionaries with the following fields:

• query - Paraphrased query written by annotators.
• docs - List of relevant document titles.
• original_query - The original query which was paraphrased. Atomic queries are enclosed by <mark></mark>. Augmented queries do not have this field populated.
• scores - This field is not populated and only used when producing predictions to enable sharing the same data structure.
• metadata - A dictionary with the following fields:
  • template - The template used to create the query.
  • domain - The domain to which the query belongs.
  • fluency - List of fluency ratings for the query.
  • meaning - List of ratings for whether the paraphrased query meaning is the same as the original query.
  • naturalness - List of naturalness ratings for the query.
  • relevance_ratings - Dictionary mapping document titles to relevance ratings for the document.
  • evidence_ratings - Dictionary mapping document titles to evidence ratings for the document.
  • attributions - Dictionary mapping a document title to its attributions attributions are a list of dictionaries mapping a query substring to a document substring.

The document corpus is at https://storage.googleapis.com/gresearch/quest/documents.jsonl. Note that this file is quite large (899MB). The format is newline separated json dicts containing title and text.

Citation Information

@inproceedings{malaviya23expertqa,
    title = {QUEST: A Retrieval Dataset of Entity-Seeking Queries with Implicit Set Operations},
    author = {Chaitanya Malaviya and Peter Shaw and Ming-Wei Chang and Kenton Lee and Kristina Toutanova},
    booktitle = {ACL},
    year = {2023},
    url = "https://arxiv.org/abs/2305.11694"
}