Hugging Face's logo

Datasets:
ericmacedo
/
LitSearch2k

Tasks:
Text Retrieval
Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Size:
1K - 10K
Tags:
research
litsearch
hard-negatives
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
corpusid
int64
74.1k
265M
title
large_stringlengths
0
247
abstract
large_stringlengths
0
8.73k
citations
listlengths
0
68
full_paper
large_stringlengths
1.3k
136k
token_count
int64
260
32.2k
6,791,168
Automatic Text Simplification for Spanish: Comparative Evaluation of Various Simplification Strategies
In this paper, we explore statistical machine translation (SMT) approaches to automatic text simplification (ATS) for Spanish. First, we compare the performances of the standard phrase-based (PB) and hierarchical (HIERO) SMT models in this specific task. In both cases, we build two models, one using the TS corpus with ...
[ 12093312, 13603486, 8884845, 809364, 9945908, 15636533, 38407095, 1395225, 18594619, 5219389, 4896510 ]
Automatic Text Simplification for Spanish: Comparative Evaluation of Various Simplification Strategies Sep 7-9 2015 Sanjaštajner sanjastajner@wlv.ac.uk Research Group in Computational Linguistics University of Wolverhampton UK Iacer Calixto icalixto@computing.dcu.ie ADAPT Centre School of Computing Dublin City Univ...
9,162
260,063,238
Knowledge-Augmented Language Model Prompting for Zero-Shot Knowledge Graph Question Answering
Large Language Models (LLMs) are capable of performing zero-shot closed-book question answering tasks, based on their internal knowledge stored in parameters during pre-training. However, such internalized knowledge might be insufficient and incorrect, which could lead LLMs to generate factually wrong answers. Furtherm...
[]
Knowledge-Augmented Language Model Prompting for Zero-Shot Knowledge Graph Question Answering July 13, 2023 Jinheon Baek jinheon.baek@kaist.ac.kralham.fikri@mbzuai.ac.aeamsafari@amazon.com Alham Fikri Aji Amir Saffari Kaist Mbzuai Amazon Knowledge-Augmented Language Model Prompting for Zero-Shot Knowledge Graph...
23,601
254,877,704
WeCheck: Strong Factual Consistency Checker via Weakly Supervised Learning
"A crucial issue of current text generation models is that they often uncontrollably generate text t(...TRUNCATED)
[4711425,233289483,207853069,3432876,244345901,233204406,218571335,204960716,236477519,239009616,207(...TRUNCATED)
"WeCheck: Strong Factual Consistency Checker via Weakly Supervised Learning\nLong PapersCopyright Lo(...TRUNCATED)
27,189
237,420,553
Data Augmentation for Cross-Domain Named Entity Recognition
"Current work in named entity recognition (NER) shows that data augmentation techniques can produce(...TRUNCATED)
[1671874,49577956,52967399,1845735,3257353,225041226,220047370,59523656,222141002,220045358,17479970(...TRUNCATED)
"Data Augmentation for Cross-Domain Named Entity Recognition\nAssociation for Computational Linguist(...TRUNCATED)
10,815
247,447,562
Hyperlink-induced Pre-training for Passage Retrieval in Open-domain Question Answering
"To alleviate the data scarcity problem in training question answering systems, recent works propose(...TRUNCATED)
[ 173990818, 211068995, 215737187, 3618568, 26501419, 10910955, 6401679, 86611921, 231704318 ]
"Hyperlink-induced Pre-training for Passage Retrieval in Open-domain Question Answering\nLong Papers(...TRUNCATED)
13,421
259,076,105
Finding the SWEET Spot: Analysis and Improvement of Adaptive Inference in Low Resource Settings
"Adaptive inference is a simple method for reducing inference costs. The method works by maintaining(...TRUNCATED)
[ 238856994, 208117506, 233189542, 215785895, 214802887, 5034059, 3432876, 11816014, 233296246, 990233 ]
"Finding the SWEET Spot: Analysis and Improvement of Adaptive Inference in Low Resource Settings\nLo(...TRUNCATED)
13,755
259,833,868
Can Large Language Models Safely Address Patient Questions Following Cataract Surgery?
"Recent advances in large language models (LLMs) have generated significant interest in their applic(...TRUNCATED)
[]
"Can Large Language Models Safely Address Patient Questions Following Cataract Surgery?\nJuly 14, 20(...TRUNCATED)
6,399
221,970,445
TernaryBERT: Distillation-aware Ultra-low Bit BERT
"Transformer-based pre-training models like BERT have achieved remarkable performance in many natura(...TRUNCATED)
[52967399,214802887,6628106,8451212,3323727,49667762,202750230,2753399,202888986,162183964,13239389,(...TRUNCATED)
"TernaryBERT: Distillation-aware Ultra-low Bit BERT\n\n\nWei Zhang zhangwei379@huawei.com \nHuawei N(...TRUNCATED)
13,578
250,150,926
CAMS: An Annotated Corpus for Causal Analysis of Mental Health Issues in Social Media Posts
"Research community has witnessed substantial growth in the detection of mental health issues and th(...TRUNCATED)
[ 233189632, 220047810, 53137220, 49208337, 247594449 ]
"CAMS: An Annotated Corpus for Causal Analysis of Mental Health Issues in Social Media Posts\n\n\nMu(...TRUNCATED)
12,516
254,017,551
"Competency-Aware Neural Machine Translation: Can Machine Translation Know its Own Translation Quali(...TRUNCATED)
"Neural machine translation (NMT) is often criticized for failures that happen without awareness. Th(...TRUNCATED)
[245855939,215744964,216642180,91184134,52967399,13751870,218487046,52100616,207847180,245855921,153(...TRUNCATED)
"Competency-Aware Neural Machine Translation: Can Machine Translation Know its Own Translation Quali(...TRUNCATED)
11,564
End of preview. Expand in Data Studio

LitSearch 2k: A Hard-Negative Research Corpus

This dataset is a curated 2,000-document subset of the LitSearch corpus, designed specifically for testing retrieval robustness against sophisticated "hard negatives."

Evaluation Statistics

  • Total Documents: 2,000 (574 Gold, 1,426 Distractors)
  • Mean Hard Negatives in Top-20 (Dense): 6.9
  • Queries with Zero Hard Negatives: 0
  • Mean Distractors per Query: 2.39

Dataset Description

The corpus was constructed using a multi-phase subsampling pipeline that prioritizes documents which are lexically similar, semantically close, or bibliographically related to the gold standard results, while explicitly excluding the gold documents themselves (except where they are part of the target evaluation set).

Selection Criteria

  1. Lexical Traps: Documents with high BM25 scores but low semantic similarity to the query.
  2. Semantic Proximity: Documents within the same embedding clusters as gold results (E5-large-v2).
  3. Citation Neighbors: Direct forward and reverse citations of gold documents.
  4. Retrieval Bleed: Documents that consistently appear in the top-k results across multiple queries.
  5. Metadata Overlap: Documents sharing authors or venues with the gold standard set.

Dataset Structure

Features

  • corpusid: Unique identifier for the paper (consistent with S2ORC/LitSearch).
  • title: The title of the paper.
  • abstract: The paper abstract.
  • full_paper: The extracted text of the paper.
  • is_gold: Boolean flag indicating if this document is a gold standard result for at least one query in the 2k subset.
  • year: Publication year.

Usage

This dataset is intended for use in IR research, specifically for evaluating "needles in haystacks" scenarios and the effectiveness of re-ranking models on extremely challenging distractor sets.

from datasets import load_dataset

# 1. Load the document corpus (default config)
corpus = load_dataset("ericmacedo/LitSearch2k")

# 2. Load the queries and gold standards (queries config)
queries = load_dataset("ericmacedo/LitSearch2k", "queries")

Citation

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

@inproceedings{ajith-etal-2024-litsearch,
    title = "{L}it{S}earch: A Retrieval Benchmark for Scientific Literature Search",
    author = "Ajith, Anirudh  and
      Xia, Mengzhou  and
      Chevalier, Alexis  and
      Goyal, Tanya  and
      Chen, Danqi  and
      Gao, Tianyu",
    editor = "Al-Onaizan, Yaser  and
      Bansal, Mohit  and
      Chen, Yun-Nung",
    booktitle = "Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.emnlp-main.840/",
    doi = "10.18653/v1/2024.emnlp-main.840",
    pages = "15068--15083",
}

Bibtex for the dataset:
@misc{litsearch2k_dataset,
  author       = {Cabral, Eric},
  title        = {LitSearch 2k: A Hard-Negative Retrieval Subsampling Pipeline for LitSearch},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/ericmacedo/LitSearch2k},
  version      = {1.0.0}
}
Downloads last month
115
Number of rows:
2,581
Total file size:
52.4 MB