README.md

metadata

license: apache-2.0
task_categories:
  - information-retrieval
  - text-retrieval
tags:
  - beir
  - nq
  - information-retrieval
  - retrieval
  - search

BEIR NQ Dataset (Migrated)

This is a migrated version of BeIR/nq that is compatible with datasets library 4.0.0+.

Dataset Description

This dataset contains the nq dataset from the BEIR benchmark, converted from the old script-based format to Parquet format.

Dataset Structure

Queries

• Split 'queries': 3,452 examples

  • Features: ['_id', 'text', 'metadata']

• Total examples: 3,452

Corpus

• Split 'corpus': 2,681,468 examples

  • Features: ['_id', 'title', 'text', 'metadata']

• Total examples: 2,681,468

Usage

from datasets import load_dataset

# Load queries (split: queries)
queries = load_dataset("Hyukkyu/beir-nq", "queries", split="queries")

# Load corpus (split: corpus)
corpus = load_dataset("Hyukkyu/beir-nq", "corpus", split="corpus")

Available Splits

Queries

• queries: 3,452 examples

Corpus

• corpus: 2,681,468 examples

Original Dataset

This dataset is migrated from: BeIR/nq

Citation

If you use this dataset, please cite the original BEIR paper:

@article{thakur2021beir,
  title={BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of Information Retrieval Models},
  author={Thakur, Nandan and Reimers, Nils and Ruckle, Andreas and Srivastava, Abhishek and Gurevych, Iryna},
  journal={arXiv preprint arXiv:2104.08663},
  year={2021}
}