BEIR Quora (orgrctera/beir_quora)

Overview

This release packages the Quora Duplicate Questions slice of the BEIR (Benchmarking IR) benchmark as a flat, table-oriented dataset for retrieval evaluation and tooling. The underlying content comes from Quora’s public Question Pairs resource: pairs of user questions labeled as duplicate or not. BEIR reformulates the problem as duplicate question retrieval: each query is one question text, and systems must rank a large corpus of Quora questions so that semantically duplicate questions (the “relevant” documents for that query) appear at the top.

The task is community Q&A: questions are informal, short, and highly paraphrastic—lexical overlap between a query and its duplicates is often weak, which stresses semantic retrieval and re-ranking.

In the BEIR benchmark suite, Quora is categorized as duplicate-question retrieval; the corpus is large (on the order of hundreds of thousands of question texts in the standard BEIR distribution), with multiple gold duplicates possible per query (BEIR reports roughly 1.6 relevant items per query on average in the benchmark statistics).

This Hub dataset is a tabular export aligned with other CTERA benchmark releases: one row per query with expected_output as a JSON string of relevant corpus document IDs (BEIR qrels-style supervision).

Task

• Task type: Retrieval for Quora duplicate questions (dense/sparse retrieval, re-ranking, or full RAG-style pipelines over the BEIR Quora corpus).
• Input (input): A natural-language question (the retrieval query).
• Reference (expected_output): A JSON string listing relevant corpus document IDs with scores (typically binary 1 for duplicate / relevant), e.g. [{"id": "243211", "score": 1}, ...]. Evaluators retrieve against the full BEIR Quora corpus (distributed with BEIR; not duplicated in every row) and score with standard IR metrics (nDCG@k, MRR, Recall@k, etc.).
• Metadata: Original BEIR query_id and split name are preserved.

Background

Quora Question Pairs (source data)

Quora released Question Pairs data to support research on identifying duplicate questions—a practical problem for Q&A sites that want to merge redundant threads. The data were popularized in the Kaggle “Quora Question Pairs” competition (2017) and are described in Quora’s dataset announcement and related community materials. Pairs are labeled for semantic equivalence (duplicate vs. not), which maps naturally to training and evaluating paraphrase and semantic similarity models.

• Community / competition context: Kaggle — Quora Question Pairs
• Quora data release note: First Quora Dataset Release: Question Pairs

BEIR reformulation

BEIR (Thakur et al., 2021) standardizes many retrieval datasets into a common corpus + queries + qrels layout so that zero-shot and cross-domain comparisons of lexical, sparse, dense, and re-ranking systems are straightforward. For Quora, each question in the corpus is a “document”, and queries are other questions; relevance indicates duplicate / paraphrase clusters. This turns pairwise duplicate detection into a full retrieval benchmark over a large candidate pool.

• Code / resources: UKPLab/beir
• Related Hub layout (corpus shards): BeIR/quora · BeIR/beir-corpus

Data fields

Column	Type	Description
id	string	Stable UUID for this row in this Hub release.
input	string	Query text (a Quora question).
expected_output	string	JSON array of {"id": "<corpus-doc-id>", "score": 1} objects — relevant (duplicate) question IDs in the BEIR corpus.
metadata.query_id	string	BEIR Quora query identifier.
metadata.split	string	dev or test.

Splits

Split	Rows
dev	5,000
test	10,000
Total	15,000

Examples

Illustrative rows as stored (IDs and text from the dataset).

Example 1 — single relevant document (dev)

• input: Is sleep ethical?
• expected_output:

[{"id": "243211", "score": 1}]

• metadata.query_id: 243212 · metadata.split: dev

Example 2 — multiple relevant duplicates (test)

• input: What is really the point of life?
• expected_output (excerpt):

[
  {"id": "192963", "score": 1},
  {"id": "264401", "score": 1},
  {"id": "48872", "score": 1},
  {"id": "136318", "score": 1},
  {"id": "54554", "score": 1}
]

• metadata.split: test

References

BEIR (benchmark)

Abstract (arXiv:2104.08663): Existing neural information retrieval (IR) models have often been studied in homogeneous and narrow settings, which has considerably limited insights into their out-of-distribution (OOD) generalization capabilities. To address this, and to facilitate researchers to broadly evaluate the effectiveness of their models, we introduce Benchmarking-IR (BEIR), a robust and heterogeneous evaluation benchmark for information retrieval. We leverage a careful selection of 18 publicly available datasets from diverse text retrieval tasks and domains…

• BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models — arXiv:2104.08663
• NeurIPS 2021 Datasets & Benchmarks (OpenReview)

Quora Question Pairs (task and community release)

The primary citation for academic use often points to the Kaggle competition and Quora’s dataset documentation; many papers refer to the task as paraphrase / duplicate detection over Quora question pairs.

• Kaggle — Quora Question Pairs

Citation

If you use the BEIR benchmark formulation, cite:

@inproceedings{thakur2021beir,
  title={{BEIR}: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models},
  author={Thakur, Nandan and Reimers, Nils and R{\"u}ckl{\'e}, Andreas and
          Srivastava, Abhishek and Gurevych, Iryna},
  booktitle={Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2)},
  year={2021},
  url={https://openreview.net/forum?id=wCu6T5xFjeJ}
}

Provenance

Exported for retrieval evaluation (e.g. Langfuse / internal tooling) with Quora as the BEIR sub-benchmark quora. Corpus question texts are not inlined in each row; join expected_output document IDs to the BEIR Quora corpus when building an index.