datalama/miracl-hard-negatives

MIRACL Hard Negatives (Parquet Format)

This is a Parquet-converted version of mteb/miracl-hard-negatives, compatible with the latest HuggingFace datasets library (4.0+).

Why This Dataset?

The original mteb/miracl-hard-negatives uses a Python script-based loader, which is no longer supported in datasets >= 4.0.0. This dataset provides the same data in standard Parquet format.

Dataset Description

MIRACL (Multilingual Information Retrieval Across a Continuum of Languages) is a multilingual retrieval dataset that focuses on search across 18 different languages.

The hard negatives version was created by pooling the top 250 documents per query from:

• BM25
• e5-multilingual-large
• e5-mistral-instruct

This makes the retrieval task more challenging compared to the standard MIRACL dataset.

Languages

Code	Language
ar	Arabic
bn	Bengali
de	German
en	English
es	Spanish
fa	Persian
fi	Finnish
fr	French
hi	Hindi
id	Indonesian
ja	Japanese
ko	Korean
ru	Russian
sw	Swahili
te	Telugu
th	Thai
yo	Yoruba
zh	Chinese

Usage

from datasets import load_dataset

# Load English data
corpus = load_dataset("datalama/miracl-hard-negatives", "en-corpus", split="corpus")
queries = load_dataset("datalama/miracl-hard-negatives", "en-queries", split="queries")
qrels = load_dataset("datalama/miracl-hard-negatives", "en-qrels", split="dev")

print(f"Corpus: {len(corpus)} documents")
print(f"Queries: {len(queries)} queries")
print(f"Qrels: {len(qrels)} relevance judgments")

Data Format

Queries ({lang}-queries)

Column	Type	Description
_id	string	Query ID
text	string	Query text

Corpus ({lang}-corpus)

Column	Type	Description
_id	string	Document ID
title	string	Document title
text	string	Document text

Qrels ({lang}-qrels)

Column	Type	Description
query-id	string	Query ID
corpus-id	string	Document ID
score	int	Relevance score

Citation

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

@article{zhang2022miracl,
  title={MIRACL: A Multilingual Retrieval Dataset Covering 18 Diverse Languages},
  author={Zhang, Xinyu and Thakur, Nandan and Ogundepo, Odunayo and Kamalloo, Ehsan and Alfonso-Hermelo, David and Li, Xiaoguang and Liu, Qun and Rezagholizadeh, Mehdi and Lin, Jimmy},
  journal={arXiv preprint arXiv:2210.09984},
  year={2022}
}

License

Apache 2.0 (same as the original dataset)

Acknowledgments

• Original dataset: mteb/miracl-hard-negatives
• MIRACL benchmark: miracl.ai
• MTEB benchmark: mteb