Dataset Viewer
Auto-converted to Parquet
query-id
stringclasses
280 values
corpus-id
stringlengths
2
5
score
int64
1
1
q0
d0
1
q0
d1
1
q0
d2
1
q0
d3
1
q0
d4
1
q1
d5
1
q1
d6
1
q1
d7
1
q1
d8
1
q1
d9
1
q2
d10
1
q2
d11
1
q2
d12
1
q2
d13
1
q2
d14
1
q3
d15
1
q3
d16
1
q3
d17
1
q3
d18
1
q3
d19
1
q4
d20
1
q4
d21
1
q4
d22
1
q4
d23
1
q4
d24
1
q5
d25
1
q5
d26
1
q5
d27
1
q5
d28
1
q5
d29
1
q6
d30
1
q6
d31
1
q6
d32
1
q6
d33
1
q6
d34
1
q7
d35
1
q7
d36
1
q7
d37
1
q7
d38
1
q7
d39
1
q8
d40
1
q8
d41
1
q8
d42
1
q8
d43
1
q8
d44
1
q9
d45
1
q9
d46
1
q9
d47
1
q9
d48
1
q9
d49
1
q10
d50
1
q10
d51
1
q10
d52
1
q10
d53
1
q10
d54
1
q11
d55
1
q11
d56
1
q11
d57
1
q11
d58
1
q11
d59
1
q12
d60
1
q12
d61
1
q12
d62
1
q12
d63
1
q12
d64
1
q13
d65
1
q13
d66
1
q13
d67
1
q13
d68
1
q13
d69
1
q14
d70
1
q14
d71
1
q14
d72
1
q14
d73
1
q14
d74
1
q15
d75
1
q15
d76
1
q15
d77
1
q15
d78
1
q15
d79
1
q16
d80
1
q16
d81
1
q16
d82
1
q16
d83
1
q16
d84
1
q17
d85
1
q17
d86
1
q17
d87
1
q17
d88
1
q17
d89
1
q18
d90
1
q18
d91
1
q18
d92
1
q18
d93
1
q18
d94
1
q19
d95
1
q19
d96
1
q19
d97
1
q19
d98
1
q19
d99
1
End of preview. Expand in Data Studio

📚 Translated LONG2RAG (MTEB-Style Retrieval Dataset)

Dataset Summary

This dataset is a translated version of the LONG2RAG benchmark (Qi et al., EMNLP Findings 2024), adapted into MTEB-style retrieval format for evaluating multilingual retrieval-augmented generation (RAG) and long-context retrieval systems.

LONG2RAG was originally designed to evaluate how well large language models (LLMs) incorporate key points from retrieved long documents into long-form answers. It includes 280 complex, practical questions across 10 domains and 8 question categories, each paired with 5 retrieved documents (avg. length ~2,444 words).

This translated version preserves the structure but reformats it into query–document relevance pairs suitable for retrieval evaluation under the Massive Text Embedding Benchmark (MTEB).


Supported Tasks and Leaderboards

  • Task Category: Retrieval
  • Task: Given a natural language query, rank candidate documents by relevance.
  • MTEB Integration: Compatible with mteb evaluation framework.

Languages

  • Original: English
  • This release: Translated into Persian

Dataset Details

Queries

  • 280 complex, uncontaminated, long-form questions.

Corpus

  • Retrieved real-world documents (5 per query).

Relevance Labels

  • Binary (relevant / not relevant).

Domains and Question Categories

Domains (10)

  • AI
  • Biology
  • Economics
  • Film
  • History
  • Music
  • Religion
  • Sports
  • Technology
  • Others

Question Categories (8)

  • Factual
  • Explanatory
  • Comparative
  • Subjective
  • Methodological
  • Causal
  • Hypothetical
  • Predictive

Data Splits

  • test: 280 queries

Each query has 5 candidate documents, aligned with MTEB retrieval style.


Citation

@inproceedings{qi2024long2rag,
  title = {LONG2RAG: Evaluating Long-Context \& Long-Form Retrieval-Augmented Generation with Key Point Recall},
  author = {Qi, Zehan and Xu, Rongwu and Guo, Zhijiang and Wang, Cunxiang and Zhang, Hao and Xu, Wei},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2024},
  year = {2024}
}
Downloads last month
76